Artificial Intelligence Review (2025) 58:365

https://doi.org/10.1007/s10462-025-11366-9

A dynamic operation room scheduling DORS strategy based

on explainable AI and fuzzy interface engine

Rana Mohamed El-Balka1 · Noha Sakr1 · Asmaa H. Rabie1 · Ahmed I. Saleh1

Accepted: 14 August 2025

© The Author(s) 2025

Abstract
Poor surgical scheduling causes major problems in hospital operating rooms, such as long
patient wait times, underutilized operating rooms, and high costs. Existing scheduling
approaches, which are static or less adaptable, fail to handle real-time unpredictability.
To overcome these constraints, this study presents Dynamic Operation Room Scheduling
(DORS), a new intraday surgical scheduling system. DORS uses a two-layered architec-
ture: (1) Explainable AI for feature selection that is based on critical scheduling criteria
such as Round Robin, and (2) a dynamic scheduling system that includes a Receiving
Module, a Checking Module for patient prioritization, and a Scheduling Module provided
by a Fuzzy Interface Engine. This system allows for proactive schedule preparation and
reactive modifications, making it possible to smoothly include unscheduled surgical opera-
tions. In comparison to traditional (FCFS, Round Robin) and optimization-based (genetic
algorithm) methods. DORS dynamically modifies schedules to reduce average wait times
(AWT), consistently outperforming other approaches by 120–560 min. DORS completes
surgical operations more quickly (half of surgical operations in 255–725 min). In addition,
DORS retains a modest runtime (45 ms) while increasing scheduling efficiency (98.6%).
DORS also demonstrates strong stability, with low Relative Percentage Deviation (RPD)
on high-demand days. Finally, DORS achieves the optimal blend of speed, efficiency, and
responsiveness, making it the greatest choice for hospitals aiming to eliminate delays,
optimize operating room usage, and effectively manage changing surgical needs.

Keywords Explainable AI · Fuzzy interface · Machine learning · Operating room ·

Patient scheduling · Surgery scheduling

Rana Mohamed El-Balka

[email protected]
1
Computer and Control Systems Engineering Department, Faculty of Engineering, Mansoura
University, Mansoura, Egypt

13
 365 Page 2 of 39 R. M. El-Balka et al.

1 Introduction

A healthcare system is a structural organization that provides people treatment, medical

services, and preventive measures. Hospitals play an essential role in health systems, and
operating rooms (ORs) are the expensive services because they utilize a large amount of
valuable resources, including personnel, equipment, and surgeons, accounting for between
35 and 40% of total expenses (Bellini et al. 2020). Furthermore, ORs directly impact patient
safety and are the main source of annual revenue (Liu et al. 2011). Therefore, any improve-
ments in OR efficiency have the potential to increase hospital productivity and improve
patient satisfaction (Durán et al. 2017). As a result, integrated planning and scheduling
may be the key to hospital management and efficiency improvement (Eshghali et al. 2024).
Operating room (OR) performance is improved from many angles, and a wide range of
approaches is suggested. Nonetheless, many of these projects aimed at enhancing the OR
don’t appear to make it into reality (Choudhary et al. 2024). Suggested modifications occa-
sionally fail to address the intended issue or may not work well with the OR’s overall work-
flow. The OR is a complicated setting with intricate social relationships, unpredictability,
and a low tolerance for mistakes (Etherington et al. 2021). From a financial and patient care
quality perspective, careful planning and resource allocation remain essential. By reducing
patient wait times, OR idle time, and overtime, a well-designed appointment system can
increase productivity (Lotfi and Behnamian 2022). The choice of surgical operations to be
performed, the allocation of OR resource time, and the sequence of procedures within the
allotted time are all aspects of the surgery scheduling problem. Some problems in the OR
are triggered by a combination of factors such as demanding caseloads, pressure to perform
complex procedures, and conflicting priorities. Therefore, healthcare workers may experi-
ence more mental strain and stress as a result (Harris and Claudio 2015). Optimizing the
OR is high on the academic agenda due to the effects of the workflow on patients, pressure
on healthcare professionals working in the OR, the dynamic nature of the workplace, and
financial restrictions. However, it is challenging to manage and alter the system because of
the high standards set for patients, the interactions between many experts, the unpredictable
nature of the surgical case scheduling, and so on (Schouten et al. 2023).
Unscheduled emergencies account for 40% of OR outages, which increases idle time
by 18%. Hospitals that have dedicated emergency operating rooms minimize elective case
delays by 30%. For example, (Shetty et al. 2023) suggested an intraday dynamic resched-
uling model that changes forthcoming appointments in response to observed no-shows to
calculate the ideal pre-day schedule and the optimal approach for updating the schedule in
the event of no-shows. An estimate of the surgery time may vary greatly based on many
variables, including the surgical technique, the patient’s physical state, the expertise of the
surgeon, the quantity of support personnel on hand, the time of the treatment, and the kind of
anesthesia used (Dexter and Epstein 2024). When scheduling surgery, we take into account
the most relevant information, including surgeon availability, stuff, patient status, and surgi-
cal priority. Every surgical case still requires an OR staff member who is competent in that
particular sort of operation, and aside from the surgeon, OR managers may even compose
the OR staff immediately since several operators (mainly nurses) share the required skills.
However, a shortage of nurses qualified to handle certain surgical situations in a timely
manner might cause delays and necessitate rescheduling the surgical operations. Hence,
parameters are a factor influencing surgical procedures (Abdalkareem et al. 2021).

13
A dynamic operation room scheduling DORS strategy based on… Page 3 of 39 365

This paper presents a new efficient operating room scheduling system called Dynamic
Operation Room Scheduling (DORS), which consists of two layers. The first is feature
selection by Explainable AI with scheduling criteria, and the second layer contains three
modules: Receiving Module (RM), Checking Module (CM), and Scheduling Module (SM).
In the Receiving Module (RM) and Checking Module (CM), patients are received, and who
the patient will be for surgery is determined. In the Scheduling Module (SM), fuzzy logic
is used to obtain the final schedule of surgical operations. Important issues include dynamic
rescheduling for emergencies, avoiding surgeon/nurse over-allocation, and providing timely
patient care. This paper’s contributions are summed up as follows:

● New strategy: Dynamic Operation Room Scheduling (DORS).

● Dynamic Operation Room Scheduling (DORS) consists of two layers [pre-processing
and scheduling engine modules].
● The pre-processing layer determines the effect of each parameter (feature) on surgical
procedures using Explainable AI (SHAP) with scheduling criteria.
● The scheduling engine modules layer contains three modules: real-time status input
(Receiving Module), multiple-criteria priority evaluation (Checking Module), and
fuzzy rule-based improvement (Scheduling Module).
● New proposal based on the fuzzy interface engine to order surgical operations in sched-
uling.
● Dealing with out-of-hospital emergency patients more quickly and getting them into
the schedule.
● Determining whether and for how long surgery can be postponed based on parameters.

This paper is organized as follows: Sect. 2 presents the previous efforts in operating room
scheduling. Section 3 focuses on the proposed dynamic operation scheduling strategy. Sec-
tion 4 depicts the experimental results. The conclusions are discussed in Sect. 5.

2 Literature review

The progress in medical diagnosis and treatment is increasing daily. However, this advance-
ment also brings new challenges, such as lengthy patient queues at medical facilities and
labs, which lead to a staffing shortages relative to patient volume and subpar service. In this
section, traditional patient scheduling methods are presented alongside new approaches to
healthcare schedule planning. These methods focus and reducing wait times, optimizing
resource utilization, managing patient delays, enhancing efficiency, distributing workloads,
controlling healthcare costs, and prioritizing patient preferences. In (Xiao and Yoogalingam
2022), Xiao et al. introduce different options for scheduling outpatient surgical operations in
the context of urgent arrivals. They first use simulation optimization to find heuristic solu-
tions for a surgery scheduling problem involving a heterogeneous mix of urgent and elec-
tive operations in a multi-operating room setting. Then, they statistically evaluate different
sequencing and allocation rules using a discrete-event simulation model.
In (Ala et al. 2021), Ala et al. demonstrate the use of the whale optimization algorithm
(WOA), based on the NSGA-II algorithm and the Pareto archive, to solve an appointment
scheduling model while incorporating simulation. Roziqin et al. (2020) propose that soft-

13
365 Page 4 of 39 R. M. El-Balka et al.

ware information systems can help health services refer patients to intended hospitals more
promptly and accurately, emphasizing the need for continuous development to meet public
demand.
In (Alharbi and AlQahtani 2016), Alharbi et al. provide a genetic algorithm solution to
the scheduling issue facing the physicians at Prince Sultan Military Medical City (PSMMC)
in Riyadh, Saudi Arabia’s pediatric department. A cost bit matrix is used in the genetic algo-
rithm technique, where each cell represents a constraint violation. Lin et al. (2020) propose
that within a week, a redesigned mathematical model be built to allocate procedures to
operating rooms. They suggest four simple heuristics that may be used efficiently. More-
over, four local search techniques that might greatly enhance a given answer are discussed.
The issue under study is solved using a hybrid genetic algorithm (HGA) that combines local
search techniques, elite search techniques, and early solutions.
In (Kianfar and Atighehchian 2023), they present combining the linear programming and
simulated annealing metaheuristic models to provide a new heuristic method and mixed-
integer programming model. Sensitivity analysis is performed on factors, such as the cost of
operating room overtime, the maximum amount of permitted overtime, the number of beds
available in the ward, and the number of days each surgeon would want to be present. In
(Lin and Yen 2023), it is suggested to use a genetic algorithm (GA) to solve the scheduling
issue for operating rooms. The suggested GA’s effectiveness is assessed by the testing of
randomly created issue scenarios. de Quiroz et al. (2023) provide a variety of techniques to
solve the issue, utilizing both inner metaheuristic components and simpler re-optimization
heuristics as well as more intricate scenario-based strategies. To increase the effectiveness
of the ED, they also investigate the impact of the number of physicians and the potential for
early patient information.
Yuniartha et al. (2023) assess the effect of duration prediction on the performance of
operating room scheduling using a category value. They provide a scheduling model for
the operating room that takes into account the preferences of surgeons regarding the start
time of the procedure while managing restricted resources. The model is solved through the
use of a priority dispatching rule-based heuristic method. Boccia et al. (2024) introduced a
condensed and effective integer linear programming (ILP) formulation that integrates the
20-min tight responsiveness requirement into offline planning for emergency management
for the integrated operating room planning and scheduling (IORPS) problem. They offer a
study of the produced solutions that comply with hospital specifications and may be applied
to improve emergency management system responsiveness and efficiency. In (Tsang, et
al. 2024), Tsang et al. propose solving the following decision-making problems simulta-
neously: an assignment problem that designates an OR and an anesthesiologist for each
surgery; an allocation problem that chooses which ORs to open and which on-call anesthe-
siologists to call in; and a sequencing and scheduling problem that establishes the sequence
of surgeries and the times at which they should begin in each OR. In (Azar et al. 2022), it
is suggested to use chance constrains based on the probability distribution of the surgery
length for each surgeon to enhance scheduling performance. They demonstrate how to add
these chance constraints to time-indexed formulations as linear ones.
In (Ala and Goli 2024) a mixed-integer linear programming (MILP) paradigm is
described. Furthermore, it presents a realistic technique that uses the combined machine
learning-tabu search (ML-TS) method to reduce overtime expenses in operating rooms
while increasing patient satisfaction. In (Ala 2024) a technique is provided for scheduling

13
A dynamic operation room scheduling DORS strategy based on… Page 5 of 39 365

each surgeon’s waiting list based on days spent in one of the operating rooms. Two objec-
tives are considered: reducing costs related to overtime and unutilized operating rooms.
The proposed solution comprises two approaches: mathematical modeling and optimization
using simulation-based methodologies. The simulation-based optimization solution matches
the quality of the mathematical model’s answer for smaller issues while providing a quick
and adequate solution for larger-scale problems. In (Ala et al. 2024), a unique paradigm is
suggested for improving treatment standards in smart healthcare systems (SHS) through
the confluence of AI and IoT, in addition to improving patient data processing performance
in smart healthcare facilities. The enhanced particle swarm optimization-long short-term
memory (PSO-LSTM) technique is used to improve the IoT-based SHS model. To conduct a
more appropriate categorization of patient medical data, PSO is compared with PSO-LSTM
to modify numerous metrics and benchmarks to attain the best value on patient data process
performance.
In (Merghani et al. 2025), the review illustrates the importance of machine learning in
detecting surgical case cancellations, optimizing post-anesthesia unit resource allocation,
and predicting surgical case length. Neural networks, XGBoost, and random forests are
examples of machine learning algorithms that have demonstrated promise in prediction
accuracy and resource efficiency. The study stresses how machine learning is evolving in
surgical surgery and how more innovation is needed to properly achieve AI’s transforma-
tional promise for patients, healthcare workers, and clinicians. Overall, using AI in OR
administration has the potential to improve patient outcomes and healthcare productivity.
In (Park et al. 2025), it is suggested to improve the accuracy of these projections beyond
current estimation methods by developing random forest models tailored to certain surgical
departments. A huge dataset is used to test a variety of machine learning algorithms, includ-
ing Random Forest, XGBoost, Linear Regression, LightGBM, and CatBoost. Additionally,
SHAP-based feature importance is used. Department-specific random forest models are
implemented to enhance surgical scheduling by providing a more accurate and dependable
tool for anticipating surgical case durations. The related studies are summarized in Table 1.

3 The proposed dynamic operation scheduling strategy (DORS)

The operating room (OR) scheduling challenge involves optimizing the allocation of pro-
cedures to available ORs while balancing efficiency, resource constraints, and patient pri-
ority. To minimize delays, decrease costs, and maximize OR usage, existing hospital and
emergency patients must be managed under conditions of unpredictability, such as varying
operation times, personnel availability, and equipment limits. Advanced systems (such as
fuzzy logic in the proposed DORS system) resolve ambiguity by ranking instances based
on urgency, resource availability, and risk, allowing for adaptive decision-making in real
time. The objective is to create a cost-effective, patient-centered schedule that enhances
hospital performance and staff satisfaction. The proposal consists of two layers. In the first
layer, the data is processed by identifying the features that have the greatest impact on surgi-
cal scheduling, using scheduling criteria such as First Come, First Served, Short Job First,
Round Robin, etc. The feature impact on the scheduling criteria output is then calculated
using explainable AI, and the less important features are excluded. In the second layer,
the Scheduling Engine Modules is divided into three modules: Receiving Module (RM),

13
 365 Page 6 of 39 R. M. El-Balka et al.

Table 1 Comparison of the latest scheduling methods

References Year Scheduling Techniques Limitations Benefits Sched- Dataset
type uling
time
Alharbi and 2021 Doctors Software infor- Without Simple and Daily Jember
AlQahtani Scheduling mation system optimization flexible public
(2016) and may be un- hospital
able to handle
complicated
restrictions
Lin and 2016 Genetic In large-scale Manages shift Par- Pediatric
Chou algorithm issues, con- changes very ticular Depart-
(2020) vergence may well days ment of
take longer than and Prince
expected shifts Sultan
Military
Medi-
cal City
(PSMMC)
in Riyadh/
Saudi
Arabia
Lin and 2023 Combining sim- Real-time im- Balances Par- Teaching-
Yen (2023) ulated annealing provements are between ex- ticular education-
meta-heuristic computation- ploration and days al hospital
and linear ally costly extraction
programming
Kianfar and 2020 Patient Hybrid ge- GA environ- High-quality Week Belgian
Atighehchi- Scheduling netic algorithm ments need to university
an (2023) (HGA) be optimized hospital
Yuniar- 2023 Metaheuristic The method Flexible Daily Two hospi-
tha et al. needs more tals (Hong
(2023) details Kong SAR
of China
and Italy)
Boccia et 2023 Rule-based heu- Not appropri- Fast and Daily Middle-
al. (2024) ristic method ate for big recognizable scale
or dynamic university
situations hospital in
Indonesia
Tsang et al. 2024 Linear Challenges Accurate and Daily Naples
(2024) programming with nonlinear optimum hospital
or stochastic under
restrictions limitations
Azar et al. 2024 Stochastic pro- Computational- Manages Daily Data from
(2022) gramming (SP) ly complicated Uncertainty a health
and distribu- system in
tionally robust New York
optimization
(DRO) models
Ala (2024) 2024 – Two Approach Difficult High efficien- Hour- Multi-
(mathematical implementation cy and quick- ly− hospital
modeling and ness + Re- Daily EHR + OR
optimization sponds to manage-
through sim- emergency ment
ulation-based situations systems
methods)

13
A dynamic operation room scheduling DORS strategy based on… Page 7 of 39 365

Table 1 (continued)
References Year Scheduling Techniques Limitations Benefits Sched- Dataset
type uling
time
Ala et al. 2024 – Particle swarm Depends on High accu- Real Smart
(2024) optimization- GPU resources racy, real-time time Healthcare
long short- and difficult analysis and Systems
term memory implementation more secure (iot + HER)
(PSO-LSTM) data

1 Pre-processing

Features
Features Scheduling criteria Explainable Ai importance

2 Scheduling Engine Modules

Receiving Module Checking Module Scheduling Module

Fig. 1 The proposed dynamic operation scheduling strategy (DORS) framework

Checking Module (CM), and Scheduling Module (SM). Figure 1 shows the framework for
the proposal (DORS).

3.1 Pre-processing layer based on explainable AI and scheduling criteria

In data processing, there are two main methods: Round Robin and explainable AI (XAI)
(Ali et al. 2023). Initially, surgical operations are scheduled based on features using the
Round Robin method. Then, the scheduling of surgical operations is input into explainable
AI, which outputs the relative importance of each feature in the surgical scheduling process.
Explainable AI, based on the most recent developments, has two techniques that work with
a range of data and model types and attempt to explain the model outputs: SHapley Addi-
tive exPlanations (SHAP) (Elkhawaga et al. 2023) and Local Interpretable Model-Agnostic
Explanation (LIME) (Hung and Lee 2024).
SHAP is a game theory-based XAI technique. By treating every feature (or predictor) as
a player and the model outcome as the reward, it seeks to explain every model. SHAP can
describe the role of the features for both a particular instance and for all instances since it
offers both local and global explanations. SHAP computes feature attribution by evaluating
various combinations. While LIME offers only local explanations, SHAP provides both

13
 365 Page 8 of 39 R. M. El-Balka et al.

global and local explanations. Additionally, while LIME fits a local linear model and is
unable to capture nonlinear relationships, SHAP may be able to do so depending on the
model used. LIME creates a single plot for each instance, while SHAP produces multiple
plots that report the results both locally and globally (Salih et al. 2024).

3.2 Scheduling engine modules

In this section, the Scheduling engine is the core part of the proposed DORS. Figure 2 shows
the DORS modules.

3.2.1 Receiving module (RM) and checking module (CM)

Receiving Module (RM) to determine the total number of patients in the hospital who will
undergo surgery. Checking Module (CM) determine if the patients are ready for surgery by
checking the patients status; if the status is not ready, the surgery will be postponed; if the
status is ready, the availability of the medical staff is checked; if they are not available, the
surgery is postponed; if available, the surgery will be listed for scheduling.

3.2.2 Scheduling module (SM)

In the third stage, we receive the surgical operations that are scheduled to take place during
the day, and each surgery contains several parameters whose values differ from one surgery
to another, which changes the priority of each surgery. However, with several parameters
and their values, it becomes very difficult to determine the priority of each surgery. Based
on that, fuzzy logic is used to determine the priority of each surgery so that we can schedule
the surgical operations. After scheduling the surgical operations and executing a surgical
operation, a surgical operation may arrive from outside the schedule that was set (opi).
Therefore, in the event that a new surgical operation (opA) arrives, it is examined to see if
it is an emergency or not. In the event that it is not an emergency, it is added to the surgical
operations that were scheduled and rescheduled again. But if it is an emergency, we check
to see if the operation currently being executed (opi) can be hold or not to see if it is permis-
sible to hold it to allow the new surgical operation (opA) to be executed. We have two direc-
tions: the first path is if the operation currently being executed cannot be hold then redirect
(opA) and continue the current operation; when finished, fetch the next surgical operation
from the scheduling of the surgical operations. The second path, if the operation currently
being executed can be to hold, then execute the new surgical operation (opA). When (opA)
is finished, continue the (opi) surgery. But if the new surgical operation (opA) is not finished
and another new surgical operation (opA) arrives, we check if the new surgical operation is
an emergency. If Yes, then redirect (opA), if No, continue the first (opA).
Designing a Fuzzy Inference Engine for Operating Room (OR) scheduling entails many
critical phases, including identifying input and output variables, selecting acceptable mem-
bership functions, creating a comprehensive rule base, and selecting a suitable defuzzifica-
tion approach. This section elaborate on each of these points and explain how the fuzzy
inference engine can deal with the uncertainty inherent in OR scheduling. The Fuzzy Inter-
face Engine is based only on simple IF–THEN rules merged with fuzzy logic operators (e.g.,
AND and OR) to enhance the decision-making, which is very similar to humans reasoning.

13
A dynamic operation room scheduling DORS strategy based on… Page 9 of 39 365

Start Checking Module (CM)

Patient status
Not
Receiving Module (RM) Ready
Ready

Stuff Availability? No Postponing surgery

1 2 3 4 5 6
5
Opi

Scheduling Module (SM)

Execute operation (Opi)

No

1 2 3 4 6
New Operation (OpA)
Yes
Fuzzy Interface Check if (OpA) is
emergency?
Fuzzificaon Defuzzificaon Yes

Fuzzy Rules Check

No No
Holding of
Add to (Opi)?
Yes
Schedule query
Redirect (OpA)
3 4 2 Execute operation (OpA)
6 1
Continue (Opi)
New Operation (OpA)? Yes
No
No
No
(OpA) Finished
No
Yes Check (Opi)
Check if (OpA) is Yes finished?
emergency?

Is there another No
Redirect (OpA)
Yes surgery?
Yes
Fetch next op
End

Fig. 2 A flowchart demonstrating the control flow across the three different modules of the proposed
scheduling engine

Fuzzy inference is a process to map the input into an output using fuzzy logic (Gamel et
al. 2022). The process goes as follows: (1) The input of crisp values is converted into fuzzy
quantities, (2) it goes through the fuzzy rules and fuzzy memberships to generate an output,
and (3) the output is in the form of a fuzzy set that needs to be defuzzified to get the output
of parameter values once again. The input to the fuzzy system is the parameter values such
as Time (T), Availability (A), Priority (P), and Status (S) from Table 7, where we have 9
surgical operations for each parameter. They need to be converted into fuzzy sets using the

13
 365 Page 10 of 39 R. M. El-Balka et al.

memberships shown in Fig. 3. Membership functions denote the degree to which a crisp
input value belongs to a fuzzy set. The membership function shape and parameters used
have a considerable influence on the functioning of the fuzzy inference engine. Triangular
membership functions are frequently used for their ease of use and computational effec-
tiveness. To generate the corresponding memberships for the four parameters, we need to
determine α, β, and ϒ values as presented in Eq. 1 (Qaid et al. 2023).
∑n
0.5× (value)x
α= x=1
n , β = α × 2, Υ = α × 3 (1)

After calculating the three values of α, β , and Υ, each value from each ranking is con-
verted into fuzzy input that can be small (S), medium (M), or large (L) by using Eqs. 2 to
4. Figure 3 illustrates a graphical representation of Eqs. 1 to 4 of the different considered
membership functions.
{
1x ≤ α
µ(x)small = β−x
β−α α< x ≤ β (2)
0x > β

 0x ≤ α
 x−α
<x≤β
β−α α
µ(x)medium = Υ−x (3)

 Υ−β β <x≤γ
0x > γ
{
0x ≤ β
µ(x)large = x−β
Υ−β β < x ≤ γ (4)
1x > γ

Consequently, the fuzzy rule-based input is the output of the fuzzification process. A set of
rules is considered here in the form IF (X is A) AND (Y is B) THEN (Z is C), where X, Y, and
Z represent input variables (e.g., T, A, P, and S), and A, B, and C represent the corresponding
linguistic variables (e.g., small, medium, and large). The first part of the rule (before THEN)
is called “antecedent”. The second part (after THEN) is called “consequent”. These input

Fig. 3 Membership function and determine α, β, and γ using the data used in the next example

13
A dynamic operation room scheduling DORS strategy based on… Page 11 of 39 365

fuzzy sets go through the if–then rules to determine the output. In this paper, there are 81
different rules used to determine the output, as shown in Table 2. Rule formulation depends
on OR Scheduling Logic: Rules are produced using OR scheduling personnel’s expertise
and experience, as well as their specific scheduling objectives. Example rules include “IF
(Time is small) AND (Priority is large) THEN (Schedule Priority is large)”. “IF (Avail-
ability is small) OR (Status is small) THEN (Schedule Priority is small)”. “IF (Priority is
Emergency) THEN (Schedule Priority is large)”. “IF (Time is large) AND (Availability
is Medium) THEN (Schedule Priority is Medium)”. Rules may be created that take into
account the “Time” variable in addition to other aspects. For example, a long procedure
with high priority may nevertheless be scheduled ahead of a very long surgery with medium
priority. Emergency cases use specific rules with “Priority is Emergency” in the antecedent
would be assigned a “Highest Schedule Priority” in the consequent, thereby superseding
other reasons and assuring quick scheduling. If absolutely required, the “AND” operator can
be used to combine “Emergency” with other criteria (for example, resource availability);
however, in most cases, emergency status would take precedence. Then the output goes into
a defuzzification process to get crisp values back representing the final ranking.
Defuzzification can be applied using different methods such as max–min, max criterion,
center-of-gravity (COG), and the mean of maxima (El-Balka et al. 2022). The max–min
depends on choosing a min operator for the conjunction in the premise of the rule as well
as for the implication function and the max aggregation operator (El-Balka et al. 2022).
Consider a simple case of two elements of evidence per rule; the corresponding rules will be
 
aggregation
  
µM = max  min
 µkj1 , µkj2 ∀j ∈ {1, 2, 3, . . . , N }(5)
implication

This yields to:

µM = max (min (µk11 , µk12 ) , min (µk21 , µk22 ) , . . . , min (µkN 1 , µkN 2 )) (6)

COG is the most popular method, and this is the method used in the current study. This
method is identical to the formula for calculating the center of gravity in physics. The
membership function, in our case, is bounded by the weighted average of the membership
function or the COG of the area. Defuzzification can be accomplished using the output
membership function. Assuming α = 3, β = 6, and Υ = 9 related to Eqs. 1. In Algorithm
1, the steps of the fuzzy interface engine to get final scheduling are provided.

13
 365

13
Table 2 The adopted 81 fuzzy rules are used to determine the output
ID T A P S Rule output ID T A P S Rule output ID T A P S Rule output ID T A P S Rule output
Page 12 of 39

1 S S S S S 22 S L M S S 42 M M M L M 62 L S L M L
2 S S S M S 23 S L M M M 43 M M L S M 63 L S L L L
3 S S S L M 24 S L M L L 44 M M L M M 64 L M S S S
4 S S M S S 25 S L L S L 45 M M L L M 65 L M S M M
5 S S M M M 26 S L L M L 46 M L S S S 66 L M S L L
6 S S M L S 27 S L L L L 47 M L S M M 67 L M M S M
7 S S L S M 28 M S S S S 48 M L S L L 68 L M M M M
8 S S L M S 29 M S S M S 49 M L M S M 69 L M M L M
9 S S L L M 30 M S S L S 50 M L M M M 70 L M L S L
10 S M S S S 31 M S M S S 51 M L M L M 71 L M L M M
11 S M S M M 32 M S M M M 52 M L L S L 72 L M L L L
12 S M S L S 33 M S M L M 53 M L L M M 73 L L S S L
13 S M M S M 34 M S L S S 54 M L L L L 74 L L S M L
14 S M M M M 35 M S L M M 55 L S S S M 75 L L S L L
15 S M M L M 36 M S L L L 56 L S S M S 76 L L M S L
16 S M L S S 37 M M S S M 57 L S S L L 77 L L M M M
17 S M L M M 38 M M S M M 58 L S M S S 78 L L M L L
18 S M L L L 39 M M S L M 59 L S M M M 79 L L L S L
19 S L S S M 40 M M M S M 60 L S M L L 80 L L L M L
20 S L S M S 41 M M M M M 61 L S L S L 81 L L L L L
21 S L S L L
R. M. El-Balka et al.
A dynamic operation room scheduling DORS strategy based on… Page 13 of 39 365

Algorithm 1 Operating room scheduling using the proposed fuzzy algorithm

3.3 DORS illustrative example

At the beginning of this section, we start with the first layer, where data with all attributes
is entered into the Round Robin system, as shown in Table 3. Surgical operations are sched-
uled based on all features for each surgical procedure, which consists of five features. The
first feature is the time of each operation. The second feature shows staff availability, pro-
viding data between 0 and 1. The closer the value is to 1, the more staff is available. The
third feature shows the priority of the surgery, with values ranging from 0 to 1. The fourth
feature is the patient’s state if stable, and the final feature is the surgeon’s age.
The Round Robin results, as shown in Table 4, are then fed into the Explainable AI
to measure the impact of the features on the scheduling output generated by the Round
Robin model, to obtain the impact of each feature and exclude the less important ones. The
explainable AI output is shown as Fig. 4.
The Explainable AI gave Feature Importance values (0.008590, 0.013263, 0.021002,
0.006712, and 0.000363), and based on that, the top features were time, availability, priority,
and state. Therefore, the Dr_age feature will be excluded, and only four features will be fed
into the second layer.
In layer two we select the patient who will be prepared for surgery. As shown in Table 5,
14 surgical operations are required to be processed for surgery before the filtration of sur-
gical operations. As can be noted, the table includes four features, which are the patient’s
Status, time of surgical operations, priority of surgery, and Availability of medical staff.
First, patients will be identified according to their Status and the Availability of the required
medical staff prior to the filtration process.

13
 365 Page 14 of 39 R. M. El-Balka et al.

Table 3 The data used as input to implement the Round Robin method
Time (T) Availability Priority Status Dr_age Time Availability Priority Status Dr_age
(A) (P) (S) (D) (T) (A) (P) (S) (D)
Op1 75 min 0.9 0.51 0.73 30 Op8 75 min 0.9 0.58 0.37 33
Op2 120 min 0.8 0.27 0.58 45 Op9 120 min 1 0.72 0.35 40
Op3 90 min 0.9 0.23 0.21 40 Op10 90 min 0.8 0.65 0.62 50
Op4 40 min 0.7 0.82 0.29 55 Op11 40 min 0.7 0.48 0.83 60
Op5 60 min 0.8 0.91 0.82 54 Op12 60 min 1 0.59 0.45 55
Op6 120 min 0.8 0.35 0.90 30 Op13 120 min 0.6 0.71 0.41 48
Op7 180 min 0.6 0.18 0.71 35 Op14 180 min 0.5 0.92 0.18 36

Table 4 The scheduling scores obtained by implementing the Round Robin method
Round Robin Op5 Op12 Op10 Op9 Op11 Op1 Op4
Value 0.7295 0.6112 0.6040 0.6039 0.5888 0.5760 0.5744
Round Robin Op6 Op8 Op13 Op14 Op2 Op7 Op3
Value 0.5363 0.5300 0.5263 0.4978 0.4733 0.4052 0.4043

Fig. 4 Feature importance evaluation using Mean Absolute SHAP value (XAI)

Table 5 Data for all the patients ready for surgical operations before applying the checking module
Time (T) Availability Priority Status Time Availability Priority Status
(A) (P) (S) (T) (A) (P) (S)
Op1 75 min 0.9 0.51 0.73 Op8 75 min 0.9 0.58 0.37
Op2 120 min 0.8 0.27 0.58 Op9 120 min 1 0.72 0.35
Op3 90 min 0.9 0.23 0.21 Op10 90 min 0.8 0.65 0.62
Op4 40 min 0.7 0.82 0.29 Op11 40 min 0.7 0.48 0.83
Op5 60 min 0.8 0.91 0.82 Op12 60 min 1 0.59 0.45
Op6 120 min 0.8 0.35 0.90 Op13 120 min 0.6 0.71 0.41
Op7 180 min 0.6 0.18 0.71 Op14 180 min 0.5 0.92 0.18

13
A dynamic operation room scheduling DORS strategy based on… Page 15 of 39 365

Table 6 The surgical operations that have the possibility to be postponed (the highlighted ones)
Time (T) Availability Priority Status Time Availability Priority Status
(A) (P) (S) (T) (A) (P) (S)
Op1 75 min 0.9 0.51 0.73 Op8 75 min 0.9 0.58 0.37
Op2 120 min 0.8 0.27 0.58 Op9 120 min 1 0.72 0.35
Op3 90 min 0.9 0.23 0.21 Op10 90 min 0.8 0.65 0.62
Op4 40 min 0.7 0.82 0.29 Op11 40 min 0.7 0.48 0.83
Op5 60 min 0.8 0.91 0.82 Op12 60 min 1 0.59 0.45
Op6 120 min 0.8 0.35 0.90 Op13 120 min 0.6 0.71 0.41
Op7 180 min 0.6 0.18 0.71 Op14 180 min 0.5 0.92 0.18

Table 7 Surgical operations after removing postponed patients, indicating the possibility of holding opera-
tions (before scheduling)
Time (T) Availabil- Prior- Sta- Hold Hold Finish Op arrive OpA OpA
ity (A) ity (P) tus time op priority time
(S)
Op1 75 min 0.9 0.51 0.73 Yes 30 min 30 min Op1A 1 30 min
Op2 120 min 0.8 0.27 0.58 No – 40 min Op2A 1 180
min
Op3 90 min 0.9 0.23 0.52 – – – – – –
Op5 60 min 0.8 0.91 0.82 Yes 25 min 35 min Op4A 1 60 min
Op6 120 min 0.8 0.35 0.90 No – 30 min – – –
Op8 75 min 0.9 0.58 0.67 – – – – – –
Op10 90 min 0.8 0.65 0.62 No – 50 min – – –
Op11 40 min 0.7 0.48 0.83 Yes 20 min 10 min Op3A 2 45 min
Op12 60 min 1 0.59 0.85 – – – – – –

In the Checking module, the patient’s status is first taken into consideration. If the
patient’s status value is greater than 0.5, the patient can be admitted to surgery. If the value
is less than that, the patient is not allowed to be admitted to surgery. Then, the availability
of the medical staff is considered. If the value is greater than 0.7, the patient is admitted to
surgery. If the value is less than that, the surgery is postponed. The surgical operations that
should be postponed are shown in Table 6, where in Op4 the value is 0.29 and in Op9 the
value is 0.35 in status, both of which are less than 0.5. In Op7, the value is 0.6 in availability,
which is less than 0.7. In Op13 and Op14, the values in status and availability are both less
than the specified values, where in Op13 the value is 0.41 in status and 0.6 in availability, and
in Op14 the value is 0.18 in status and 0.5 in availability.
After completing the filter process, we now have 9 surgical operations from 14 in Table 7.
From here, we begin the daily scheduling process for the surgeries. We have many param-
eters, all of which are important to consider when scheduling surgical operations. That’s
why we use Fuzzy logic so we can combine all the parameters and use them in scheduling.
Figure 5 shows the steps from fuzzy logic to final priority for surgical operations. In Table 8,
surgical operations are displayed in order of priority generated by fuzzy logic, along with
surgical operations that may arrive from outside and their arrival time. The “Hold” column
indicates whether the surgery can be stopped, while the “Hold Time” column specifies the
allowable waiting time. Additionally, the “Finish Op” column specifies the end time of the

13
 365 Page 16 of 39 R. M. El-Balka et al.

Fig. 5 A detailed description of the proposed fuzzy algorithm used in surgical operation scheduling

surgery. The “Op Arrive,” “OpA Emergency” and “OpA Time” columns indicate when the
surgery will arrive, whether it is an emergency, and how long it will take.
At first, it starts with OP15 until it ends, as there is no surgical operation from outside.
Then Op6 starts working and continues until it finishes. Op5 works, and while it is working,
Op4A arrives, and needs 60 min, but Op5 can be hold for only 25 min and needs 35 min to
finish. the time required to perform surgery on Op1A is greater than the hold time available
in Op5, so it is not possible to wait for Op1A, as Op5 has 35 min left to finish, which is a long
time, especially since Op1A is an emergency and cannot wait that long. Op10 starts directly
after Op5, as no Op arrives from outside, and Op10 cannot be holding, so it continues until
finished. When Op1 is working, Op2A arrives, and it is an emergency. When checking if Op1
can be holding, the situation is as follows: Op1 needs 30 min to finish, and Op1A needs 30

13
A dynamic operation room scheduling DORS strategy based on… Page 17 of 39 365

Table 8 Surgical operations after removing postponed patients, indicating the possibility of holding opera-
tions (after schedule)
Time (T) Avail- Pri- Sta- Hold Hold Finish Op OpA OpA
ability ority tus Time Op Arrive emergency Time
(A) (P) (S)
Op12 60 min 1 0.59 0.85 – – – – – –
Op6 120 min 0.8 0.35 0.90 No – 30 min – – –
Op5 60 min 0.8 0.91 0.82 Yes 25 min 35 min Op1A Yes 60 min
Op10 90 min 0.8 0.65 0.62 No – 50 min – – –
Op1 75 min 0.9 0.51 0.73 Yes 30 min 30 min Op2A Yes 30 min
Op8 75 min 0.9 0.58 0.67 – – – – – –
Op3 90 min 0.9 0.23 0.52 – – – – – –
Op11 40 min 0.7 0.48 0.83 Yes 20 min 10 min Op3A Yes 45 min
Op2 120 min 0.8 0.27 0.58 No – 40 min Op4A No 180 min

min for surgery, but Op1 can be holding 30 min, Therefore, Op1 is hold, and Op2A goes in
for surgery. After Op2A finishes, Op1 surgery is completed.
Continuing with Op8 after Op1, when Op8 finishes, Op3 starts. While the Op8 and Op3
were working, no outside surgery arrives, so they are completed at the time specified in the
schedule. After them, Op11 works. During Op3A arrives and needs 45 min, while Op11 can be
holding for only 20 min. This is not suitable for Op3A, but considering the remaining time
of Op11, we find that it is a short time, which is 10 min, so Op3A can be prepared during that
time and begin surgery directly because it is an emergency.
Op2 surgery begins based on the schedule, but during its work, the Op4A arrives. The
Op4A is not an emergency, so Op2 continues, and Op4A is inserted into the schedule after
Op2. After all surgeries are finished, we have 12 surgical operations completed during the
day, as shown in Fig. 6.

3.4 Genetic algorithm

A genetic algorithm begins with a randomly shuffled set of surgical operations. Each indi-
vidual represents a potential scheduling strategy or ranking of these surgical operations. The
starting population is a collection of such individuals, generally generated at random. The
population represents a group of potential solutions. In code, it defaults to 100, meaning
each generation will have 100 separate schedules. These individuals are rated for fitness
using the objective function. Each schedule is scored by the objective function based on
its priority, patient case, and time. The formula rewards high-priority and high-urgency
instances while minimizing time by utilizing the inverse of time. Selection chooses the
highest-performing individuals from the population. The best half (depending on fitness
score) is preserved for breeding the following generation.
This helps ensure that positive features are carried down. Crossover combines two paren-
tal schedules to produce a new one. It takes a portion of one parent’s schedule and fills in
the remainder with the other, preventing duplication. This fosters the combination of strong
features. Mutation randomly switches two surgical operations in a schedule. This happens
seldom, only 10% of the time, but it adds variance to keep the algorithm from being trapped
with similar results. The process repeats for a number of generations (100 by default), each
time producing a new population of the best solutions. After several generations, the best

13
 365 Page 18 of 39 R. M. El-Balka et al.

Fig. 6 An explanation of the entire scheduling process for all the surgical operations in a sequential flow

solution discovered is returned as the final answer. The GA algorithm is represented in algo-
rithm 2. So the summary of the parameters:

● Population Size: Number of solutions each generation, with 100 different candidate
solutions (operational rankings).
● Mutation Rate: 10% Chance of random solution changes for each child.
● Crossover Rate: parents merge to generate a child (ordered crossover). The crossover
point is determined at random from the parent list’s first and last positions.

13
A dynamic operation room scheduling DORS strategy based on… Page 19 of 39 365

Algorithm 2 The implemented genetic algorithm’ steps

4 Evaluation and results

The current section presents experiments using our proposal (DORS), the results for each
stage, their application to the dataset, and the effect of each feature on the output using
explainable AI (SHAP). The experiments are performed on Windows 11 using the Python
programming language. The used packages are NumPy, Pandas, Schedul, skfuzzy, SHAP,
and Scikit-Learn. The environment has an Intel Core i7 processor with 16 GB of RAM.

4.1 Dataset

Operating room data from Nile Hospital (​h​t​t​p​s​:​​/​/​w​w​w​​.​k​a​g​g​l​​e​.​c​o​​m​/​d​a​t​​a​s​e​t​s​​/​r​a​n​a​e​​l​b​a​l​​k​a​/​o​p​​

e​r​a​t​i​​n​g​-​r​o​o​​m​-​s​u​​r​g​e​r​y​-​d​a​t​a​s​e​t​/​d​a​t​a) is a dataset that contains all surgical operations for the
period between December 1, 2023, and December 31, 2023. It is in the form of seven col-
umns that provide daily information as follows: (1) day, (2) priority, (3) time, (4) patient
case, (5) stuff, (6) dr_age, and (7) family number. The Staff column represents the number of
medical staff in the operating room, excluding surgeons, and the Patient Condition column
indicates the stability of the patient’s condition and ensures readiness to enter the operat-
ing room. The priority column expresses emergency cases and their priority in performing
surgery. The dataset contains two files. The first file consists of 471 samples and five col-
umns. The second file contains the surgical operations coming from outside with their status
(rejected or scheduled). Samples from the modified dataset are shown in Tables 9 and 10.
The data in its current form cannot be used for the scheduling stage because the columns
have inconsistent formats and a wide range in the numerical values for priority, time, patient
case, stuff, dr_age, and family number. Therefore, it is necessary to perform a preprocessing

13
 365 Page 20 of 39 R. M. El-Balka et al.

Table 9 Samples from the used dataset containing 6 features before preprocessing
Day Priority Time Patient case Stuff dr_age Family number
1-Dec 5 90 min Stable 6 60 3
1 35 min Stable 8 55 2
2 20 min Unstable 5 55 1
7 75 min Stable 8 55 2
7 75 min Stable 9 55 2
7 75 min Stable 7 55 2
7 75 min Stable 8 50 5
7 90 min Stable 9 48 4
2 90 min Unstable 5 38 1
2-Dec 4 60 min Stable 7 40 1
7 30 min Unstable 8 40 2
7 60 min Stable 9 40 3
7 60 min Stable 7 40 3
5 60 min Stable 7 40 4
3 60 min Stable 8 40 2
4 45 min Stable 9 38 2
2 45 min Stable 7 40 4
7 60 min Unstable 9 45 3
7 40 min Stable 8 45 3
7 60 min Stable 9 60 2
7 30 min stable 8 60 3
7 60 min Stable 8 48 2
7 30 min stable 6 48 2
7 60 min Stable 8 48 2
7 45 min Stable 9 48 2
7 30 min stable 8 48 3
7 20 min Stable 7 48 3
7 30 min Unstable 8 48 2
7 30 min Stable 9 50 3
7 30 min Stable 7 50 3
3-Dec 7 30 min Unstable 7 60 1
7 45 min Stable 8 60 2
7 45 min Unstable 8 60 2
7 45 min Stable 9 60 2
7 30 min Stable 7 60 2

stage to convert the data into an appropriate numerical form. This stage converts the date
column into a number representing the day of the month. It also scales the patient case,
stuff, and priority columns to a range between 0 and 1. Table 11 shows the results after the
preprocessing stage.
After data processing, the rejected surgical operations shown in Table 10 must be
excluded, and the processed data can be scheduled with the other surgical operations. Table
12 shows a sample of the data after processing.

13
A dynamic operation room scheduling DORS strategy based on… Page 21 of 39 365

Table 10 Samples from the Day Priority Time Patient case Arrive
dataset of surgical operations
1-Dec 7 90 min Stable Rejected
arriving from outside the hospital
before preprocessing 3 60 min Stable Scheduled
6 75 min Stable Scheduled
2 45 min Stable Rejected
2-Dec 5 60 min Stable Rejected
3 45 min Stable Scheduled
6 45 min Stable Rejected
4 40 min Stable Scheduled
7 45 min Stable Rejected
2 60 min Stable Scheduled
7 30 min Stable Rejected
3-Dec 7 30 min Stable Rejected
6 45 min Stable Scheduled
5 45 min Stable Scheduled
7 45 min Stable Rejected

4.2 Round robin implementation

The Round Robin step is the beginning of the processing layer, where the entire dataset is
input, and we obtain sample of the surgical operations table, as shown in Table 13, covering
15 days. This output depends on the feature values in each operation. Round Robin rating is
a balanced scoring method in which each surgical operations (op) is scored equally across
six parameters: priority, time, patient case, Stuff, dr_age, and family number. Each param-
eter is first normalized to a 0-1 scale by divide each value by the maximum value of that
parameter across all surgical operations. Then, these normalized parameters are weighted
equally to compute a total score for each operation. The surgical operations are ranked based
on these scores, where higher scores indicating higher importance. This strategy ensures no
single criterion dominates the ranking, resulting in a fair and unbiased approach to prioritiz-
ing operations.
For example, on the first day in Table 11, we have 9 surgical operations and 6 parameters.
In the first step, all columns are normalized, and then the maximum value for each column
is determined: priority = 0.7, time = 90, patient status = 0.8, stuff = 0.9, dr_age = 60, and
family number = 5.
Consider the first operation in the table with values: Priority = 0.5, Time = 90, Patient
Status = 0.7, Stuff = 0.9, dr_age = 60, and Family Number = 3.
The normalized values are calculated as following: priority 0.5/0.7 = 0.714, time 90/90 =
1.0, patient case 0.7/0.8 = 0.875, stuff 0.9/0.9 = 1.0, dr_age 60/60 = 1.0, and family number
3/5 = 0.6. The total score is the average of these normalized values: score = (sum of normal-
ized values)/number of parameters. In the second step, the first step is repeated for all surgi-
cal, and the surgical operations are ordered from highest to lowest score.

4.3 SHAP implementation

In this section, we describe the implementation of Explainable AI (XAI) using SHAP to

calculate feature importance in the dataset. The goal is to determine the impact of each fea-
ture on the output (scheduling of surgical operations) and assess their relative importance.

13
 365 Page 22 of 39 R. M. El-Balka et al.

Table 11 Samples from the used dataset containing 6 features after features preprocessing
Day Priority Time Patient case Stuff dr_age Family number
1 0.5 90 0.7 0.9 60 3
1 0.1 35 0.5 0.8 55 2
1 0.2 20 0.3 0.5 55 1
1 0.7 75 0.8 0.8 55 2
1 0.7 75 0.5 0.9 55 2
1 0.7 75 0.7 0.7 55 2
1 0.7 75 0.7 0.8 50 5
1 0.7 90 0.5 0.9 48 4
1 0.2 90 0.4 0.5 38 1
2 0.4 60 0.5 0.7 40 1
2 0.7 30 0.3 0.8 40 2
2 0.7 60 0.9 0.9 40 3
2 0.7 60 0.6 0.7 40 3
2 0.5 60 0.5 0.7 40 4
2 0.3 60 0.8 0.8 40 2
2 0.4 45 0.5 0.9 38 2
2 0.2 45 0.8 0.7 40 4
2 0.7 60 0.3 0.9 45 3
2 0.7 40 0.5 0.8 45 3
2 0.7 60 0.8 0.9 60 2
2 0.7 30 0.5 0.8 60 3
2 0.7 60 0.9 0.8 48 2
2 0.7 30 0.7 0.6 48 2
2 0.7 60 0.7 0.8 48 2
2 0.7 45 0.8 0.9 48 2
2 0.7 30 0.6 0.8 48 3
2 0.7 20 0.7 0.7 48 3
2 0.7 30 0.4 0.8 48 2
2 0.7 30 0.6 0.9 50 3
2 0.7 30 0.7 0.7 50 3
3 0.7 30 0.7 0.8 60 1
3 0.7 45 0.6 0.9 60 2
3 0.7 45 0.6 0.7 60 2
3 0.2 45 0.6 0.7 60 2
3 0.7 30 0.5 0.7 60 2

Table 12 Samples from the Day Priority Time Patient case

dataset of surgical operations
1 0.3 60 0.7
arriving from outside the hospital
after preprocessing (excluding 1 0.6 75 0.5
rejected operations) 2 0.3 45 0.8
2 0.4 40 0.5
2 0.2 60 0.6
3 0.6 45 0.8
3 0.5 45 0.6

13
 Table 13 Scheduling surgical operations for each day using Round Robin method
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
op3 op2 op10 op6 op1 op1 op6 op1 op6 op5 op9 op13 op21 op10 op2 op3
op4 op3 op1 op7 op2 op2 op2 op3 op2 op6 op23 op10 op11 op11 op5
op5 op9 op2 op8 op3 op3 op4 op4 op6 op3 op1 op8 op1 op3 op12 op2
op6 op11 op3 op9 op10 op4 op6 op7 op1 op5 op4 op4 op4 op8 op15 op22 op3 op11 op15 op14 op1
op1 op14 op6 op11 op9 op10 op8 op9 op7 op2 op8 op9 op10 op2 op7 op16 op15 op4
op7 op16 op5 op20 op12 op10 op8 op5 op7 op11 op5 op12 op4 op17
op2 op17 op8 op16 op13 op11 op9 op11 op5 op12 op4 op2 op1 op1
op9 op4 op9 op11 op14 op5 op5 op10 op17 op9 op3 op13 op19 op3 op5 op3
op8 op8 op4 op1 op19 op8 op18 op19 op1 op16 op20 op4 op6 op4 op5
op12 op7 op13 op6 op20 op21 op14 op21 op5 op2 op6
op10 op3 op7 op14 op15 op10 op6 op7 op7
op13 op4 op11 op12 op16 op2 op11 op8 op8 op8
A dynamic operation room scheduling DORS strategy based on…

op5 op5 op12 op3 op13 op7 op12 op9 op9 op9
op6 op2 op17 op22 op17 op13 op14 op12 op10
op7 op18 op14 op15 op13 op16
op1 op11 op14 op13
op15 op16 op18 op17
op6 op18
op7 op19
op17 op20
Page 23 of 39
365

13
 365 Page 24 of 39 R. M. El-Balka et al.

The Gini coefficient, a measure of impurity used in decision tree algorithms (random
forests). A Higher Gini Coefficient for a feature indicates that it is more significant for node
splitting and tree decision-making. The Random Forest Slope model was used on the train-
ing data (X_train, y_train).
Figure 7 displays four plots, two sets of feature significance visualizations for a machine
learning model (Gini Coefficient and Mean Absolute SHAP Value). Each pair includes a
bar chart displaying global feature significance and a SHAP value summary plot displaying
instance-level feature contributions. On the left side, the feature ‘day’ has the greatest Gini
coefficient, implying that it is the most relevant factor according to this measure. Where
‘Stuff’ and ‘familynumber’ have the lowest Gini coefficients, implying that they are the least
relevant qualities according to this metric. However, on the right section, it indicates the
“SHAP value (impact on model output).” SHAP values measure each feature’s contribution
to the prediction of a single occurrence. A positive SHAP number implies that the feature’s
value increases the prediction, whereas a negative SHAP value decreases it. This represents
the “feature value.” For example, high-priority values (pink/red dots) have positive SHAP
values, indicating that increasing the priority improves the model’s output. Low priority
values (blue dots) typically have negative SHAP values, lowering output. day: The influence
of ‘day’ is more variable, with both low and high values having positive and negative effects
on production; however, there appears to be a minor trend for greater ‘day’ values to have
a positive impact. Patientcase: The influence of ‘patientcase’ is mixed. Higher ‘familynum-
ber’ values often have a negative influence. Stuff: Compared to other features, ‘Stuff’ has a
minimal influence. The distribution of dots along the x-axis for each characteristic reflects
the extent to which it influences the model’s output across distinct instances. Features with
a broader distribution have a more profound and diverse influence. The same in the other
plots in Mean Absolute SHAP Value, but the result is different.
Table 14 shows a numerical ranking of feature importance based on SHAP values using
the mean absolute SHAP value and Gini coefficients. Higher values indicate the importance
of the feature in the decision-making process of the model. So in the mean absolute SHAP
value, the time feature is the highest value, where the order of the features is (priority, day,
time, patient case, dr_age, Family number, stuff), while in the Gini coefficients, the day
feature is the highest value, where the order of the features is (day, priority, time, patient
case, dr_age, Family number, stuff). Although the order of the two methods is different, they
agree that the first four features have the highest impact, and hence these features will be
used as inputs for the fuzzy logic, and they are (priority, day, time, and patient case).

4.4 An in-depth evaluation of the DORS strategy for assessing performance

effectiveness

The current subsection deals with the implementation of the proposed method (DORS) on
the dataset. In the Receiving Module (RM), the total number of surgical operations received
is 471, and after inputting them into the Checking Module (CM), we have 412 surgical
operations that need to be scheduled on their respective days. Figure 8 shows the surgical
operations in RM and CM on each day, for example, on the first day, the output from RM is
9 surgical operations; whereas after CM, we have 7 surgical operations. Table 15 shows the
result of the dataset in the Scheduling Module (SM) before any external emergency surgical
operations are introduced.

13
A dynamic operation room scheduling DORS strategy based on… Page 25 of 39 365

Fig. 7 SHAP values for each feature based on the mean absolute SHAP value and Gini coefficients

Table 14 Feature importance val- Feature Mean absolute SHAP value Gini coefficient
ues using Mean Absolute SHAP
Priority 0.424850 0.217734
value vs. Gini Coefficients
Day 0.293213 0.366095
Time 0.200211 0.119154
Patient case 0.120563 0.106438
dr_age 0.090399 0.073966
Family number 0.058499 0.070427
Stuff 0.033530 0.046185

operations
Receiving Module (RM) Checking Module (CM)

30
27

2527
20 25

2325
Number of surgerises

2123

23

25
22
21

17 21

21

21
21
20

20
19
18

20
17

17

17
17

17
17
17

16

16

16

16
15
15

15
14
13
13

15
11

11

10
9

9
9

10
8
8
7

67
6

6
6

56
5
5

5
3
3

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Days

Fig. 8 Number of surgical operations in the Receiving Module vs. the Checking Module

13
 365 Page 26 of 39 R. M. El-Balka et al.

Table 15 The output values Day Priority Time Patient case Stuff Scheduling module
resulted from the Scheduling
1 0.5 90 0.7 0.9 8.58
Module (SM) using fuzzy for
each operation 1 0.1 35 0.5 0.8 7.45
1 0.7 75 0.8 0.8 7.4
1 0.7 75 0.5 0.9 7.45
1 0.7 75 0.7 0.7 8.68
1 0.7 90 0.7 0.8 9.62
1 0.7 90 0.5 0.9 6.37
2 0.4 60 0.5 0.7 4.48
2 0.7 60 0.9 0.9 5.44
2 0.7 60 0.6 0.7 6.13
2 0.5 60 0.5 0.7 4.83
2 0.3 45 0.8 0.8 7.23
2 0.4 45 0.5 0.9 5.01
2 0.2 40 0.8 0.7 5.35
2 0.7 60 0.5 0.8 7.17
2 0.7 30 0.8 0.9 7.11
2 0.7 60 0.5 0.8 7.17
2 0.7 30 0.9 0.8 7.15
2 0.7 60 0.7 0.8 6.35
2 0.7 45 0.8 0.9 5.24
2 0.7 30 0.6 0.8 6.74
2 0.7 20 0.7 0.7 6.22
2 0.7 30 0.6 0.9 6.24
2 0.7 30 0.7 0.7 7.17
3 0.7 45 0.7 0.8 6.65
3 0.7 45 0.6 0.9 6.74
3 0.7 30 0.6 0.7 6.34
3 0.2 30 0.6 0.7 5.28
3 0.7 180 0.5 0.7 4.95

Figure 9 shows a sample of the resulting dataset for the scheduling module (SM) and the
corresponding values for each surgical procedure. Values are displayed for four days of the
month, representing the values for each surgery on each day.
For example, on day 1, there are seven surgical operations with specific values (8.58,
7.45, 7.4, 7.45, 8.68, 9.62, and 6.37) before any additional surgical operations arrive from
outside. Based on these values, the surgery with the highest value is listed first in the table,
followed by the remaining surgical operations, from highest to lowest. The same approach
applies to the surgical operations on the other days, such as day 2, day 3, and day 4 in the
figure.
After scheduling the surgical operations, additional surgical operations arrive from out-
side and need to be inserted between the existing ones. These operations are rescheduled
and placed in the appropriate order. This is shown in Fig. 10. On the first day, there were 7
surgical operations, but after including the external operations, the total increased to 9. The
surgical operations coming from outside occupied numbers 4 and 8 after rearranging all
surgical operations based on their values. The same process applies to the remaining days,
as shown in Fig. 10.

13
A dynamic operation room scheduling DORS strategy based on… Page 27 of 39 365

2-Dec

7.23

7.17

7.17

7.17
7.15
7.11
1-Dec

6.74
6.35

6.24
6.22
6.13
8

5.44

5.35

5.24
5.01
4.83
20

4.48
6

9.62
8.68
8.58
7.45

7.45

6.37
7.4

rank
rank
10 4

0 1 2
2
3
4
5
6
7
0
operations

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
operations

4-Dec
3-Dec

7.69

7.55

7.45
10
6.74
6.65

6.22
7.1

7.1
6.34

6.34

5.89

6.7
5.71

5.48
5.44
10

5.31

5.31
5.28

5.28

5.27

6.1
5.18

5.18
4.95

4.94

4.54
4.43
7
6

6
rank

rank

5
5 5
0
0
1
2
3
4
5
6
7
8
9
10
11

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
operations
operations

Fig. 9 Sample of output dataset values for surgical operations in the Scheduling Module (SM), excluding
operations arriving from outside

2-Dec
7.23
7.17
7.17
7.17
7.15
7.11
6.74

1-Dec
6.35
6.24
6.22
6.13

8
5.44
5.35
5.24
5.01

4.83

4.48
4.9

4.5
20 6
5
9.62
8.68
8.58

rank
7.45
7.45

4
6.37
7.5

7.4
6.5
rank

10 2
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
1
2
3
4
5
6
7
8
9

operations operations

3-Dec 4-Dec
6.74

6.65

6.34
6.34
6.7

7.69
6.5

7.55
5.71

7.45

8 10
5.28
5.28
7

4.95

7.1
7.1

6.22
6.8
6.7

5.89
6
6

5.48
5.44
6.1

5.31
5.31
5.27
5.18
5.18

6
4.94
4.54
4.43
6
rank

4
rank

5
2
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

operations
operations

Fig. 10 Sample of output dataset values for surgical operations in the scheduling module (SM), including
operations arriving from outside

4.5 Performance metrics

The performance measures used are (1) average wait time and (2) throughput, which are
calculated to assess efficiency for static and dynamic surgical operations before and after the
inclusion of surgical operations arriving in from outside.

13
 365 Page 28 of 39 R. M. El-Balka et al.

4.5.1 Average wait time

The average wait time is the sum of all patients’ wait times divided by the total number of
patients. In the dataset, the “Time” column reflects the operating time in minutes. The Aver-
age Wait Time (AWT) can be calculated as Eq. (9):

Total Wait Time

AWT =  (9)
Number of operations

4.5.1.1 Throughput Throughput quantifies how much work can be accomplished in a given
amount of time. It helps monitor progress and determine how well procedures are perform-
ing. By monitoring throughput, you may make modifications to increase the efficiency of
the strategy. Tracking throughput is similar to monitoring the health of the development
process. Throughput can be calculated as Eq. (10):

WIP
Throughput = (10)
cycle time

4.5.1.2 Scheduling efficiency Scheduling Efficiency assesses how well a scheduling sys-
tem uses available resources (e.g., surgical rooms, time, personnel) to fulfill surgical opera-
tions. It expresses the ratio of productive time to total available time as a percentage. Higher
efficiency translates to decreased idle time, and greater resource usage. It can be calculated
as Eq. (11).

total time of surgeries

Scheduling Efficiency = × 100 %(11)
total operational room time

4.5.1.3 Relative percentage deviation (RPD) Relative Percentage Deviation is a statisti-

cal metric that measures how much a measured or observed value deviates from a refer-
ence value, represented as a percentage of the reference. It is used to compare strategies or
models (e.g., DORS vs. FCFS in surgical scheduling), evaluate experimental vs. theoretical
scores, and examine measurement variability or bias. It can be calculated as Eq. (12).

Xi − Xref
RPD = × 100% (12)
Xref

where Xi is the value of the metric for i strategy, Xref is the reference value.

4.5.1.4 Computational time Computational time is the time it takes for a scheduling algo-
rithm to generate an optimal or near-optimal schedule. This is particularly important in
real-time scheduling, where delays in decision-making can reduce operational efficiency.
Table 16 shows the results of the average time measured when running each algorithm on
surgical operations over a 15-day period. It is nearly instantaneous in FCFS and round robin

13
A dynamic operation room scheduling DORS strategy based on… Page 29 of 39 365

Table 16 Performance evaluation FCFS Round Robin GA DORS

of DORS vs. other strategies in
Average Computation Time 2 ms 5 ms 250 ms 45 ms
terms of average computation
time

Table 17 Average waiting time DORS FCFS Round Robin GA

(AWT) across multiple days for
AWT (1day) 230 238 239 230
DORS and other strategies (in
minutes) AWT (2 day) 318 300 319 417
AWT (3 day) 225 296 249 324
AWT (4 day) 371 465 466 462
AWT (5 day) 281 307 281 300
AWT (6 day) 215 272 250 283
AWT (7 day) 458 490 529 529
AWT (8 day) 130 157 159 150
AWT (9 day) 210 240 244 239
AWT (10 day) 376 457 399 447
AWT (11 day) 552 590 600 595
AWT (12 day) 326 330 335 340
AWT (13 day) 400 595 590 595
AWT (14 day) 320 290 325 400
AWT (15 day) 111 132 132 114

but suffers from long wait periods. GA is unsuitable for real-time applications (250 ms is
too slow for hospitals). DORS takes 45 milliseconds but improves scheduling efficiency. In
our technique, runtime is measured once each day using static scheduling (because surgical
operations are fixed). And in dynamic scheduling, runtime is measured whenever additional
surgical operations are added (to capture rescheduling delays).

4.5.1.5 A comparative analysis: DORS strategy vs. other traditional static strategies In this
subsection, the average waiting time (AWT) for each of DORS, FCFS, Round Robin, and
the genetic algorithm is calculated. From the results shown in Table 17, display 15 days of
the month, the average waiting time is calculated for each method for all surgical operations
inside the hospital per day. The proposed method achieved the lowest average waiting time
among the other methods except on the first day, when the genetic algorithms showed the
same value as the proposed method, and the fifth day, when the Round Robin matched the
proposed method. Figure 11 shows the results of average waiting time with relative percent-
age deviation. DORS is the most stable algorithm, whereas others frequently deviate signifi-
cantly. FCFS and Round Robin outperform GA at times, although GA is more consistent on
certain days (for example, Day 15). On high AWT days (for example, Day 13), all options
perform worse than DORS, resulting in the greatest variances.

Table 18 shows the throughput results as measured on half the number of surgical opera-
tions during the day. Our proposed method had the lowest time to complete half the surgical
operations during the day, reducing the chance of postponing surgical operations to the next
day. However, the results showed that the Round Robin method gave the same results as our

13
 365 Page 30 of 39 R. M. El-Balka et al.

Fig. 11 a The Average Waiting Time (AWT) corresponding to each scheduling strategy applied daily over
a two-week period after external surgical operations arrival. b The corresponding Relative Percentage
Deviation (RPD) for the scheduling strategies applied on each day over a two-week interval

DORS strategy on the 12th day, and the genetic algorithms matched our DORS strategy on
the 15th day. Figure 12 shows the result of the strategies.
After calculating average waiting time, throughput must be calculated to determine the
scheduling efficiency for each of DORS, FCFS, Round Robin, and the genetic algorithm.
From the results shown in Table 19, the total efficiency for 30 days of the month is dis-
played, and the scheduling efficiency is calculated for each method for all surgical opera-
tions inside the hospital. The results of the proposed method were the highest in efficiency
among the other methods and are illustrated in Fig. 13.

4.5.1.6 Comparative analysis: DORS strategy vs. other traditional dynamic strategies In
Table 20, 15 days of the month were shown, and the average waiting time is calculated for
each method for all surgical operations inside the hospital daily, including surgical opera-
tions that came from outside and were allowed to enter. The results of the proposed method
were the lowest in average waiting time among the other methods except for the ninth day,
when FCFS showed the same value as the proposed method. Figure 14 shows the results,

13
A dynamic operation room scheduling DORS strategy based on… Page 31 of 39 365

Table 18 Comparison of throughput metric across scheduling strategies with corresponding surgical loads
per day
GA Round FCFS DORS Half surgical op- Total surgical day
Robin erations in day operations in
day
315 315 315 275 4 7 Throughput (day 1)
415 390 390 375 9 17 Throughput (day 2)
365 255 420 240 6 11 Throughput (day 3)
360 475 560 345 10 20 Throughput (day 4)
355 350 355 335 7 13 Throughput (day 5)
335 285 275 255 6 11 Throughput (day 6)
565 565 535 530 11 22 Throughput (day 7)
160 185 185 145 3 6 Throughput (day 8)
320 300 300 270 5 9 Throughput (day 9)
555 400 525 370 9 17 Throughput (day 10)
630 690 735 515 12 23 Throughput (day 11)
480 360 375 360 8 15 Throughput (day 12)
785 780 720 500 11 21 Throughput (day 13)
405 470 330 315 9 17 Throughput (day 14)
165 195 195 165 3 5 Throughput (day 15)

where the DORS strategy is the lowest average waiting time with surgical operations that
came from outside.

Table 21 shows the Throughput results as measured by half the number of surgical opera-
tions per day after adding outside surgery. After including outpatient surgical operations,
the number of surgical operations per day increased, which affected the order of surgical
operations in all methods. However, our proposed method had the shortest time to complete
half the surgical operations per day after rescheduling and including outside surgical opera-
tions in the daily schedule on 14 days, as no outside surgical operations were scheduled on
day 15.
Figure 15 shows the result of all strategies after the addition of the surgical operations
from outside. In a comparison of scheduling algorithms, DORS is the most consistent per-
former, with the lowest average waiting times over all days. While FCFS and Round Robin
function with occasional improvements, they often have longer and more unpredictable
wait times than DORS. GA has the most different results, occasionally matching DORS’
efficiency but usually exhibiting considerable performance degradation, especially on high-
demand days such as Days 2 and 13, when its relative percentage deviation scales up to
+48%.
Table 22 shows the scheduling efficiency results after adding external surgical opera-
tions, indicating that our proposed method provides the highest efficiency, followed by the
GA method, then the Round Robin method, and finally the FCFS method. The efficiency
changed after adding external operations, as the proposed method achieved the highest effi-
ciency at 98.6. The efficiency of GA and Round Robin also increased, reaching 93.5 and
88.7 respectively, while the efficiency of FCFS decreased to 80.3. Figure 16 shows the
values of the techniques.
In Table 23, the comparison between DORS, FCFS, Round Robin, GA, and two other
studies is shown. In (Ribino et al. 2022), multi-agent reinforcement learning (MARL) is

13
 365 Page 32 of 39 R. M. El-Balka et al.

Fig. 12 a The Throughput corresponding to each scheduling strategy applied daily over a two-week pe-
riod. b The corresponding Relative Percentage Deviation (RPD) for the scheduling strategies applied on
each day over a two-week interval

Table 19 Evaluating the scheduling efficiency of DORS vs other implemented techniques in terms of total
surgical time utilized relative to the available operating room time (static)
DORS FCFS Round Robin GA
Efficiency Value 96.07% 82.2% 85.4% 91.3%

an advanced AI (Q-learning algorithm) with multiple agents to optimize the scheduling of

surgery. In (Eshghali et al. 2024), the ML-IS system employs LSTM neural networks to
forecast emergency state arrivals and reinforcement learning (RL) to proactively resched-
ule elective procedures. DORS has exceptional overall performance with 98.6% dynamic
efficiency, 120–560 min average wait times, and a computational time of 45 ms, making it
perfect for most hospitals due to its combination of flexibility, speed, and cost-effective-
ness. While MARL achieves slightly higher static efficiency (98.2%) and ML-IS provides
predictive benefits (98.2% dynamic efficiency, 125–560 min average waiting time), both
require significant resources: MARL requires expensive GPU infrastructure. FCFS and

13
A dynamic operation room scheduling DORS strategy based on… Page 33 of 39 365

scheduling efficiency
100
Value in Percentage 98
96
94
92
90
88
86
84
82
80
DORS FCFS Round Robin GA
methods

Fig. 13 Scheduling efficiency percentage values of DORS compared to FCFS, Round Robin, and GA
strategies

Table 20 Average waiting time DORS FCFS Round Robin GA

(minutes) of DORS and other
AWT (1day) 200 238 230 230
strategies after the arrival of
external surgical operations AWT (2 day) 290 300 320 400
AWT (3 day) 225 296 250 320
AWT (4 day) 400 465 480 460
AWT (5 day) 285 307 290 310
AWT (6 day) 250 272 270 292
AWT (7 day) 469 490 549 539
AWT (8 day) 150 157 165 160
AWT (9 day) 240 240 254 245
AWT (10 day) 410 457 465 450
AWT (11 day) 560 590 610 600
AWT (12 day) 320 330 335 350
AWT (13 day) 440 595 570 550
AWT (14 day) 300 290 350 420
AWT (15 day) 120 132 140 125

Round Robin, despite their 2–5 ms computational time, are ineffective for complex cases
with longer average wait times and lower efficiency (80.3–85.4%). GA is unsuitable for
real-time applications due to its 250 ms computational time. For large, highly resourceful
institutions, ML-IS provides the most sophisticated solution with its predictive capabilities,
while DORS remains the practical alternative for most clinical settings, providing excellent
stability, compliance with regulations, and staff usability at the lowest operational costs.

13
 365 Page 34 of 39 R. M. El-Balka et al.

Fig. 14 a The average waiting time (AWT) corresponding to each scheduling strategy applied daily over
a two-week period after external surgical operations arrival. b The corresponding relative percentage
deviation (RPD) for the scheduling strategies applied on each day over a two-week interval after external
surgical operations arrive

5 Conclusion

This study presents DORS, a promising approach for intraday surgical scheduling that
addresses the challenges of variable patient access, resource constraints, and the need for
efficient resource allocation. DORS outperforms existing scheduling approaches by integrat-
ing Explainable AI feature importance and a three-module architecture (Receiving Module
(RM), Checking Module (CM), and Scheduling Module (SM)) based on a fuzzy interface
engine. The key strength of DORS is its dynamic scheduling, which enables responsiveness
to real-time changes and unexpected surgical procedures, optimizes resource utilization to
reduce waiting time, improves patient satisfaction through shorter waits times and faster
access to care, and leverages data-driven decision-making using AI approaches to guide
scheduling decisions. Our proposed technique was compared to three other methods (FCFS,
Round Robin, and genetic algorithm). DORS delivers a 98.6% efficiency rate compared to

13
A dynamic operation room scheduling DORS strategy based on… Page 35 of 39 365

Table 21 A comparative analysis of throughput across various scheduling strategies for daily surgical loads
after invoking externally arriving surgical operations
GA Round FCFS DORS Half surgical Number of sur- Total day
Robin operations in gical operations surgical
day from outside operations
in day
375 375 405 320 5 2 7 Throughput (day 1)
470 415 450 430 10 3 17 Throughput (day 2)
465 300 450 285 7 2 11 Throughput (day 3)
455 595 590 390 11 2 20 Throughput (day 4)
355 355 355 350 7 1 13 Throughput (day 5)
370 285 275 260 6 1 11 Throughput (day 6)
605 640 575 570 12 2 22 Throughput (day 7)
220 295 235 205 4 2 6 Throughput (day 8)
374 354 380 324 6 2 9 Throughput (day 9)
555 400 525 370 9 1 17 Throughput (day 10)
770 810 735 725 13 2 23 Throughput (day 11)
520 400 435 375 9 3 15 Throughput (day 12)
780 760 720 510 11 1 21 Throughput (day 13)
405 450 330 335 9 1 17 Throughput (day 14)
165 195 195 165 3 0 5 Throughput (day 15)

other methods and reduces surgical waiting time to a range of 120–560 min. Furthermore,
the system achieves higher throughput, completing half of the surgical operations in a day
within 255 to 725 min. While the average DORS computation time was 45 ms, we faced
a challenge because the dataset we used was relatively small. Therefore, DORS modules
may encounter scalability challenges in multi-hospital networks where thousands of surgi-
cal operations are performed daily. DORS also prioritizes based on data parameters and may
ignore the nuanced preferences of physicians. In the future, we aim to apply our method
to larger datasets and networks to enhance the efficiency of scheduling multiple operating
rooms simultaneously more efficiently, incorporating interactive AI that enables surgeons to
dynamically adjust scheduling parameters.

13
 365 Page 36 of 39 R. M. El-Balka et al.

Fig. 15 a The Throughput corresponding to each scheduling strategy applied daily over a two-week pe-
riod after external surgical operations arrive. b The corresponding relative percentage deviation (RPD) for
the scheduling strategies applied on each day over a two-week interval after external surgical operations
arrive

Table 22 Evaluating the Scheduling efficiency of DORS vs other implemented techniques in terms of total
surgical time utilized relative to the available operating room time (Dynamic)
DORS FCFS Round Robin GA
Efficiency value 98.6% 80.3% 88.7% 93.5%

13
A dynamic operation room scheduling DORS strategy based on… Page 37 of 39 365

scheduling efficiency
100
98
Value in Percentage 96
94
92
90
88
86
84
82
80
78
76
74
72
70
DORS FCFS Round Robin GA
methods

Fig. 16 Scheduling efficiency percentage values of DORS compared to FCFS, Round Robin, and GA
strategies after inclusion of external surgical operations arrive

Table 23 An in-depth performance comparison between DORS against classical/modern scheduling tech-
niques using key evaluation metrics
DORS FCFS Round GA MARL ML-IS
Robin Ribino et Eshgh-
al. (2022) ali et al.
(2024)
Efficiency Static 96.07 82.2 85.4 91.3 98.2 97.5
Dynamic 98.6 80.3 88.7 93.5 97.8 98.2
Average waiting time 120–560 132–595 min 140– 125–600 min 130–570 125–
min 610 min 560
min
Throughput 165–725 195–735 min 195– 165–770 min 175–740 170–
min 810 min 735
min
Computational time 45 ms 2 ms 5 ms 250 ms 380 ms 150 ms
Relative percentage Less RPD High-changing Me- Unexpected Medium Less
deviation (stability) (Most dium change
stable) change
Best use case Most Speed in criti- Mini Only available Good usage Large
effective cal situation clinics for offline on large hospital
balancing scheduling hospital
to hospi-
tals, and
response to
emergency
Limitations Minimal Not effective Not Impracti- Need Need
operational for complex effec- cal in live expensive large
cost cases tive for operating room GPU number
com- environments of cases
plex
cases

13
 365 Page 38 of 39 R. M. El-Balka et al.

Author contributions All Author have the same contributions.

Funding Open access funding provided by The Science, Technology & Innovation Funding Authority
(STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Data availability No datasets were generated or analysed during the current study.

Declarations

Competing interests The authors declare no competing interests.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as
you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons
licence, and indicate if changes were made. The images or other third party material in this article are
included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material.
If material is not included in the article’s Creative Commons licence and your intended use is not permitted
by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

References
Abdalkareem ZA et al (2021) Healthcare scheduling in optimization context: a review. Health Technol
11:445–469
Ala A (2024) A simulation-based optimization evaluation of operating room in healthcare under limitation
capacity: a multi-objective approaches. Spect Eng Manag Sci 2(1):85–99
Ala A, Goli A (2024) Incorporating machine learning and optimization techniques for assigning patients to
operating rooms by considering fairness policies. Eng Appl Artif Intell 136:108980
Ala A et al (2021) Optimization of an appointment scheduling problem for healthcare systems based on the
quality of fairness service using whale optimization algorithm and NSGA-II. Sci Rep 11(1):19816
Ala A et al (2024) Enhancing patient information performance in internet of things-based smart healthcare
system: hybrid artificial intelligence and optimization approaches. Eng Appl Artif Intell 131:107889
Alharbi A, AlQahtani K (2016) A genetic algorithm solution for the doctor scheduling problem. ADVCOMP
5:9–13
Ali S et al (2023) Explainable artificial intelligence (XAI): what we know and what is left to attain trustwor-
thy artificial intelligence. Inf Fusion 99:101805
Azar M, Carrasco RA, Mondschein S (2022) Dealing with uncertain surgery times in operating room sched-
uling. Eur J Oper Res 299(1):377–394
Bellini V et al (2020) Artificial intelligence: a new tool in operating room management. Role of machine
learning models in operating room optimization. J Med Syst 44(1):20
Boccia M et al (2024) Integrated operating room planning and scheduling: an ILP-based off-line approach
for emergency responsiveness at a local hospital in Naples. Soft Comput. ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​0​.​1​0​0​7​/​s​0​0​5​
0​0​-​0​2​4​-​0​9​9​4​5​-​z​​​​
Choudhary V et al. (2024) A comprehensive review of patient scheduling techniques with uncertainty. Hand-
book of Formal Optimization, 933–953
de Queiroz TA et al (2023) Dynamic scheduling of patients in emergency departments. Eur J Oper Res
310(1):100–116
Dexter F, Epstein RH (2024) Fundamentals of operating room allocation and case scheduling to minimize
the inefficiency of use of the time. Perioper Care Oper Room Manag. ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​0​.​1​0​1​6​/​j​.​p​c​o​r​m​.​
2​0​2​4​.​1​0​0​3​7​9​​​​
Durán G, Rey PA, Wolff P (2017) Solving the operating room scheduling problem with prioritized lists of
patients. Ann Oper Res 258:395–414
El-Balka RM et al (2022) Enhancing the performance of smart electrical grids using data mining and fuzzy
inference engine. Multimedia Tools Appl 81(23):33017–33049
Elkhawaga G et al (2023) Evaluating explainable artificial intelligence methods based on feature elimination:
a functionality-grounded approach. Electronics 12(7):1670

13
A dynamic operation room scheduling DORS strategy based on… Page 39 of 39 365

Eshghali M et al (2024) Machine learning based integrated scheduling and rescheduling for elective and
emergency patients in the operating theatre. Ann Oper Res 332(1):989–1012
Etherington N et al (2021) Measuring the teamwork performance of operating room teams: a systematic
review of assessment tools and their measurement properties. J Interprof Care 35(1):37–45
Gamel SA, Saleh AI, Ali HA (2022) A fog-based traffic light management strategy (TLMS) based on fuzzy
inference engine. Neural Comput Appl. https://doi.org/10.1007/s00521-021-06525-2
Harris S, Claudio D (2022) Current trends in operating room scheduling 2015 to 2020: a literature
review. Operations research forum. Springer, Cham
Hung Y-H, Lee C-Y (2024) BMB-LIME: lime with modeling local nonlinearity and uncertainty in explain-
ability. Knowl-Based Syst 294:111732
Kianfar K, Atighehchian A (2023) A hybrid heuristic approach to master surgery scheduling with downstream
resource constraints and dividable operating room blocks. Ann Oper Res 328(1):727–754
Lin Y-K, Chou Y-Y (2020) A hybrid genetic algorithm for operating room scheduling. Health Care Manag
Sci 23:249–263
Lin YK, Yen CH (2023) Genetic algorithm for solving the no-wait three-stage surgery scheduling prob-
lem. Healthcare. 11(5)
Link to the dataset: ​h​t​t​p​s​:​​​/​​/​w​w​​w​.​k​a​g​g​l​​e​.​c​​​o​m​/​d​​a​t​a​s​e​​​t​s​/​r​a​n​​a​e​l​b​​a​​l​k​a​​/​o​p​e​r​a​​t​​i​n​g​-​​​r​o​o​​m​-​s​u​r​​g​e​r​y​-​d​a​​t​a​s​e​t​/​d​a​t​a
Liu Ya, Chu C, Wang K (2011) A new heuristic algorithm for the operating room scheduling problem. Com-
put Ind Eng 61(3):865–871
Lotfi M, Behnamian J (2022) Collaborative scheduling of operating room in hospital network: multi-objec-
tive learning variable neighborhood search. Appl Soft Comput 116:108233
Merghani AM et al (2025) The role of machine learning in management of operating room: a systematic
review. Cureus 17(2):1
Park J-B et al (2025) Development of predictive model of surgical case durations using machine learning
approach. J Med Syst 49(1):1–11
Qaid MSA et al (2023) Performance analysis of diabetic retinopathy detection using fuzzy entropy multi-
level thresholding. Measurement 216:112887
Ribino P, Di Napoli C, Serino L (2022) A multi-agent RL algorithm for single-day operating room schedul-
ing. In: Workshops at 18th international conference on intelligent environments (IE2022). IOS Press
Roziqin MC, Putra DS, Noor MS (2021) Information system for doctor practice scheduling at hospitals in
Jember district. In: The first international conference on social science, Humanity, and Public Health
(ICOSHIP 2020). Atlantis Press
Salih AM et al (2024) A perspective on explainable artificial intelligence methods: SHAP and LIME. Adv
Intell Syst. https://doi.org/10.1002/aisy.202400304
Schouten AM et al (2023) Operating room performance optimization metrics: a systematic review. J Med
Syst 47(1):19
Shetty A, Groenevelt H, Tilson V (2023) Intraday dynamic rescheduling under patient no-shows. Health Care
Manag Sci 26(3):583–598
Tsang MY et al. (2024) Stochastic optimization approaches for an operating room and anesthesiologist sched-
uling problem. Operations research
Xiao Y, Yoogalingam R (2022) A simulation optimization approach for planning and scheduling in operating
rooms for elective and urgent surgeries. Oper Res Health Care 35:100366
Yuniartha DR et al (2023) Adapting duration categorical value to accommodate duration variability in a next-
day operating room scheduling. Inf Med Unlocked 37:101180

Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

13