Numerical analysis of a fractional-order giving up

smoking model by using arti cial neural network

scheme
Muhammad Sadaqat Talha

COMSATS University Islamabad

Muhammad Waseem
COMSATS University Islamabad

Article

Keywords: Mathematical smoking model, Diseased model, Neural networks, Numerical computing

Posted Date: November 15th, 2024

DOI: https://doi.org/10.21203/rs.3.rs-5232616/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.

Numerical analysis of a fractional-order giving up smoking model
by using artificial neural network scheme
Muhammad Sadaqat Talha and Muhammad Waseem

E-mail addresses:
[email protected] (MS Talha),
[email protected] (M. Waseem).
Abstract. The aim of this study is to analyze the numerical performance
of the fractional-order giving up smoking model (FO-GUSM) by developing a
framework for computation by using stochastic Levenberg-Marquardt back-
propagation artificial neural networks (SLMB-ANN). The GUSM is classi-
fied into four categories, potential smokers P (t), occasional smokers L(t),
chain smoker S(t), and quit smoker Q(t). Computations are performed by
SLMB-ANN to solve the four numerical variations. Using stochastic struc-
tured LMB-ANNs, the results obtained from GUSM were presented with
training, validation, and testing processes to reduce the mean squared error
(MSE) values compared to the reference (data-driven outcomes). To assess
the efficiency, accuracy, capability, and proficiency of the suggested compu-
tational framework LMB-ANNs, a comprehensive analysis is conducted by
analyzing correlations, mean square error (MSE), state transition data, error
histograms, and regression analysis. The importance and value of the LMB-
ANNs method is confirmed by the comparison of the results, achieving an
accuracy within 5 to 7 decimal places in solving the GUSM.
Key Words: Mathematical smoking model, Diseased model, Neural net-
works, Numerical computing

1 Introduction
The increasing mortality rates among tobacco users worldwide underscore
the urgent need for global control of this widespread addiction. Many in-
dividuals initiate smoking during youth, often unaware of its hazardous im-
plications. The global tobacco epidemic represents a prominent risk factor
for non-communicable diseases like diabetes, cardiovascular conditions, heart
diseases, and cancers presenting a substantial public health challenge on a
global scale.To curb the proliferation of smoking, it is essential to recognize it
as a significant global issue. Regular exposure to smoking advertisements can
sway smokers to reconsider their habits. Cigarette smoking leads to a grad-
ual and agonizing demise and is responsible for a range of fatal illnesses.The
media plays a crucial role by disseminating valuable information about the
consequences of smoking. It can directly convey messages promoting smoking

1
cessation, thereby influencing individuals’ perceptions, attitudes, and behav-
iors towards smoking. Several comprehensive programs have been proposed
to encourage smokers to attempt quitting as part of efforts to achieve the
overarching objective of reducing tobacco use in the long run. The effec-
tiveness of mass media campaigns in promoting smoking cessation may vary
based on their intensity and duration. Mathematical models have tradition-
ally been employed to forecast the transmission of infectious diseases over
time. These models have evolved alongside advancements in understanding
communicable diseases. It’s important to highlight that the dependability
of mathematical models relies on the precision of their underlying assump-
tions and hypotheses when utilized to analyze real-world systems. Therefore,
healthcare strategists can utilize them to improve decision-making, and plan-
ning. In [1, 2, 3, 4], multiple uses of mathematical modeling are described.
Examining the diverse impact of maternal smoking during pregnancy on the
weight of newborn babies, taking into account variations in maternal age
and estimating quantile treatment effects, consistently shows a negative im-
pact of maternal smoking on low birth weight in infants, see [5]. Likewise,
the worldwide issue of smoking motivates substantial efforts to reduce its
occurrence. One study tackles this by investigating fractional-order smoking
model employing an iterative approach that integrates domain discretization
and the short memory principle, see [6]. In modern times, public policies
addressing social issues increasingly utilize media campaigns to inspire indi-
viduals. In alignment with this approach, the authors present and study a
nonlinear mathematical model designed to explore the efforts through media
to help people quit smoking. The model’s equilibria are identified, and their
stability is examined, see [7]. In [8], authors developed a smoking model
that integrates continuous age structure within the chain smokers category.
This study investigates both local and global stabilities to establish precise
threshold dynamics within the model. Numerical simulations are conducted
to analyze the dynamic outcomes and assess the effectiveness of an optimal
control strategy proposed in the study. Overall, this research enhances our
understanding of the complexities involved in smoking cessation dynamics
and provides insights into effective control measures for potential smokers.
A comparable study on a model addressing co-abuse of smoking and heroin
can be found in [9]. In this study, researchers examine the dynamics and
consequences of smoking-heroin co-occurrence in workplace settings. They
establish a nonlinear dynamical system and perform a qualitative analysis.
Another research investigates how smoking influences the risk of lung cancer
and cardiovascular diseases by developing a nonlinear compartmental model.
This model evaluates the effectiveness of media campaigns in reducing smok-
ing rates and considers relapse among former smokers.The findings under-

2
score the importance of media campaigns in raising awareness, the rapid
spread of information in quitting smoking, and the value of early interven-
tions and counseling to deter individuals from adopting harmful habits, see
[10]. To address and diminish smoking rates, the authors [11] introduce a
delayed quitting smoking model that incorporates factors related to relapse.
To address the difficulties in reducing substance abuse within communities,
a study incorporated four control variables into its framework. The optimal
control system for these variables was derived using Pontryagin’s maximum
principle, as detailed in [12]. In [13], the nonlinear dynamics of a smok-
ing model are analyzed using ANNs integrated with both global heuristic
methods and local search techniques. These examples serve as motivating
and inspiring factors for authors to explore neural networks for addressing
smoking-related model systems. The novelty of the presented research is
outlined as follows:

• The implemented ANNs utilize the Adams dataset to simulate various

transmission rate values (α) for the smoking model.
• Achieving close alignment with results from Adams’ dataset enhances the
credibility and utility of the developed ANNs for solving the smoking model.
• Caputo fractional-order derivatives (CFD) have been employed to achieve
solutions that are more representative for addressing the FO-GUSM.
• For the first time, a stochastic structure based on LMB-ANNs has been
designed for solving the FO-GUSM numerically.
• The accuracy of structure of the SLMB-ANNs, is confirmed by comparing
the proposed numerical solutions with reference solutions.
• Comparative analyses involving regression metrics, error histograms, mean
square error, and correlation metrics bolster the effectiveness of the proposed
ANNs in tackling the smoking model.
When modeling the dynamics of smoking cessation, our primary focus is
on the population. Let’s assume that N (t) represents the total population
size at any time t. The population N (t) is divided into following four cate-
gories: the first category is Potential smokers P (t), the second is Occasional
smokers L(t), the third one is Chain smoker S(t), and the category is Quit
smokers Q(t). The equations that govern the proposed model can be found
in [12]. Figure 1 provides visual representations of the GUSM, and equation
1 presents the general form of the GUSM.

3
 Figure 1: Graphic illustrations of the GUSM


dP (t)

 dt
= λ − ζ𭟋 (L(t), P (t)) − (d + ϖ) P (t), P (0) = h1
dL(t)

= ζ𭟋 (L(t), P (t)) − (℘ + d + ϖ) L(t), L(0) = h2

dt
dS(t) (1)

 dt
= ℘L(t) − (ℑ + d + ϖ) S(t), S(0) = h3
dQ(t)

= ℑS(t) − (ϖ + d) Q(t), Q(0) = h4 ,

dt

The cross-immunity in the exposed form is represented by σ over a unit

of time in the infectious subpopulation. The initial conditions (ICs) for the
system described in equation 1 are denoted as: h1 , h2 , h3 , and h4 . Moreover,
𭟋(P, L) is a function known as the uptake function, which defines the rela-
tionship between the incidence of occasional, and potential smokers. In this
study, we present the interaction between occasional, and potential smokers
as follows:
2P L
𭟋(P, L) = . (2)
P +L
Furthermore, the function specified in equation 2 denotes the harmonic mean
of occasional, and potential smokers. In several studies, the uptake function
𭟋(P, L) is frequently defined as the product of occasional, and potential
smokers, referred to as the bilinear incidence rate, as discussed in [14]. In [15],
the authors investigated the dynamics of a smoking cessation model, which
includes an uptake function defined as the square root of P and L. Moreover,
it is interesting to contemplate this uptake function as 𭟋(P, L) = P2P+LL
. Given
that P and L are non-negative quantities, it is established that the following
relationship is valid.
2P L √ P +L
≤ PL ≤ .
P +L 2

4
The mean of two numbers is a evaluation of central tendency in a dataset.
Moreover, the geometric mean is often used to rates of change or average
ratios. Unlike arithmetic or geometric means, the harmonic mean is less
influenced by a few large values and is sometimes used for variables with
significant skewness. Readers interested in biological sense related to the
harmonic mean are directed to consult references [12], [16], and [17]. The
subsequent sections of the paper include:
• The development of the FO-GUSM is described in the second section.
• The use of LMB-ANNs in a stochastic context is introduced in the third
section.
• The technique for addressing the smoking model with the created ANNs is
described in the fourth section, along with crucial justifications.
• In the fifth section, the findings are elaborated in a manner that paves the
way for future research.

2 Design of FO-GUSM
Fractional calculus, introduced centuries ago by Newton, has grown in im-
portance rapidly in the last few decades. The concepts of fractional order
derivatives and integrals were introduced to handle situations more complex
than homogeneous forms. Fractional calculus is playing its important role in
various fields of science. The derivative and integral operators of fractional
calculus are commonly used in the study of the structure and diffusion pro-
cesses of multilayer media. Many different scientists have not only studied
complex and difficult problems of fractional and integer orders using com-
puter software tools and various important techniques, but also played an
important role in solving them, see [18, 19, 20, 21]. Many mathematicians
have used fractional order derivatives in nonlinear differential models to ob-
tain numerical and analytical results, as we can clearly see in references
[22, 23, 24]. The fractional order form of GUSM is presented as below:
 α
d P (t)
= λ − ζ𭟋 (L(t), P (t)) + (d + ϖ) P (t), P (0) = h1
 dαdtL(t)

 α

= ζ𭟋 (L(t), P (t)) − (℘ + d + ϖ) L(t), L(0) = h2
dtα
dα S(t) (3)
= ℘L(t) − (ℑ + d + ϖ) S(t), S(0) = h 3
 dαdt

 α
Q(t)

dtα
= ℑS(t) − (ϖ + d) Q(t), Q(0) = h4 .

Here, the values of fractional order have been chosen from the range of [0, 1],
and α represents the fractional order derivative. The FO-GUSM is defined
using the (CFD) as shown in system 1. This study aims to solve the FO-
GUSM using AI techniques in conjunction with LMB-ANNs [25, 26].

5
3 The structured stochastic design: LMB-
ANNs
In this section, we highlight the important developments in establishing the
computational framework of stochastic LMB-ANNs to find a solution of each
variation of the FO-GUSM. Figure 2 gives a step-by-step workflow illustra-
tion provides based on a stochastic computing scheme to find a solution of
the FO-GUSM. The first section of Figure 2 illustrates the model’s repre-
sentation, the mathematical representations of the system are provided in
the second section. The third section outlines the operational processes of
the LMB-ANNs, and the final part of Figure 2 details the simulation re-
sults. The initial step displays the data set comprising various versions of
the FO-GUSM, derived from reference solutions (Adams method) used to
find a solution of the differential model (implemented through the ’NDSolve’
function in the Mathematica software). In this instruction, ”Adam” is uti-
lized with default settings for parameter execution and stopping tolerances.
The statistics for the FO-GUSM are generated by comparing the results
with those from previous studies. Furthermore, detailed information about
the reference solutions, one can see [27, 28, 29].
Figure 3 illustrates the second segment of the stochastic paradigm LMB-
ANNs, is described in a generalized context depends upon the model of single-
neuron. In first part of Figure 3, the development of the multiple layer neural
networks is depicted, while in the second part of Figure 3, the workings of the
FO-GUSM are illustrated. The devised LMB-ANNs is executed using ’nftool’
the Matlab command to adjust parameters such as hidden neurons, testing
metrics, learning strategies, and validation data. The system’s training is
conducted using the LMB scheme with backpropagation to optimize the Ja-
cobian matrix ′ jT ′ , adjusting the bias of the variables T and minimizing the
Mean Squared Error (MSE) through weight regulation. The adjustment of
each decision variable via LMB is depicted as:
je = jT x E,
jj = jT x jT,
dT = −(I x mu
je
+ jj)
,

where I and E denote the identity matrix and error. Additionally, essen-
tial theoretical aspects regarding LMB algorithms and their implementation
process can be found in references [30, 31, 32].

6
4 Simulations accompanied by discussion and
exploration
In this section, we use numerical methods to solve three different versions of
the FO-GUSM using LMB-ANNs. The results of these numerical computa-
tions are presented based on chosen values λ = 0.25, ζ = 0.006, ϖ = 0.08,
d = 0.00004, ℘ = 0.02, and ℑ = 0.000274. The initial conditions are h1 = 1.1,
h2 = 1.2, h3 = 1.7, h4 = 2.5.
Case 1: Let’s consider the FO-GUSM written as:
 d0.6 P (t)

 dt0.6
= 0.25 − 0.006W (L(t), P (t)) + (0.00004 + 0.08) P (t), P (0) = 1.1
 d0.6 L(t)

dt0.6
= 0.006W (L(t), P (t)) − (0.02 + 0.00004 + 0.08) L(t), L(0) = 1.2
d0.6 S(t)
 0.6 = 0.02L(t) − (0.000274 + 0.00004 + 0.08) S(t), S(0) = 1.7
 d0.6dtQ(t)


dt0.6
= 0.000274S(t) − (0.08 + 0.00004) Q(t), Q(0) = 2.5,
(4)
Case 2: Let’s consider the FO-GUSM shown as:
 d0.7 P (t)

 dt0.7
= 0.25 − 0.006W (L(t), P (t)) + (0.00004 + 0.08) P (t), P (0) = 1.1
 d0.7 L(t)

dt0.7
= 0.006W (L(t), P (t)) − (0.02 + 0.00004 + 0.08) L(t), L(0) = 1.2
d0.7 S(t)
 0.7 = 0.02L(t) − (0.000274 + 0.00004 + 0.08) S(t), S(0) = 1.7
 d0.7dtQ(t)


dt0.7
= 0.000274S(t) − (0.08 + 0.00004) Q(t), Q(0) = 2.5,
(5)
Case 3: Let’s say the FO-GUSM is expressed as:
 d0.8 P (t)

 dt0.8
= 0.25 − 0.006W (L(t), P (t)) + (0.00004 + 0.08) P (t), P (0) = 1.1
 d0.8 L(t)

dt0.8
= 0.006W (L(t), P (t)) − (0.02 + 0.00004 + 0.08) L(t), L(0) = 1.2
d0.8 S(t)
 0.8 = 0.02L(t) − (0.000274 + 0.00004 + 0.08) S(t), S(0) = 1.7
 d0.8dtQ(t)


dt0.8
= 0.000274S(t) − (0.08 + 0.00004) Q(t), Q(0) = 2.5,
(6)
Case 4: Let’s suppose the FO-GUSM is formulated as:
 d0.9 P (t)

 dt0.9
= 0.25 − 0.006W (L(t), P (t)) + (0.00004 + 0.08) P (t), P (0) = 1.1
 d0.9 L(t)

dt0.9
= 0.006W (L(t), P (t)) − (0.02 + 0.00004 + 0.08) L(t), L(0) = 1.2
d0.9 S(t)
 dt0.9 = 0.02L(t) − (0.000274 + 0.00004 + 0.08) S(t),
 S(0) = 1.7
 d0.9 Q(t)

dt0.9
= 0.000274S(t) − (0.08 + 0.00004) Q(t), Q(0) = 2.5,
(7)
The effectiveness of the SLMB-ANNs framework was examined by in-
putting values in the range of [0, 1] for the GUSM using 10 neurons. The
training accuracy reached 75 percent, while the validation accuracy set at 10
percent and testing accuracy is set at 15 percent. Alternative values for these

7
 Solution through the intelligence
computing paradigms to solve the GUSM

Graphical illustrations
to represent the GUSM

Formulation of the fractional order system for

the dynamic of the four categories of the
mathematical non-linear model based smoking
by using the formation of the data set by using
the stochastic computing LMB-ANNs

Generated form of Construction of the

System of
Start The SMM by using proposed solutions
FO-GUSM through the ANNs for
The Adam’s method
GUSM

Training Standards Neurons

through the Finish
LMB method Satisfied Increased

STEP:4

Proposed outcomes using the

comparison through the
numerical reference solution
based on the absolute error
Results based on the LMB-ANNs
by using the convergence, EHs
& performances via regression
measures for each variant

Figure 2: Diagrammatic depictions of the devised LMB-ANNs workflow for

resolving the FO-GUSM: First tage provides an summary of the issue. Second
stage illustrates the mathematical system. Third stage details the practical
implementation using LMB-ANNs. Fourth stage presents the simulation re-
sults accompanied by analysis. 8
Figure 3: Systematic and versatile development of the ANNs to address the
FO-GUSM

9
 Best Validation Performance is 2.1844e-09 at epoch 513 Gradient = 9.9378e-08, at epoch 513
1010
Train

gradient
100 Validation
100
Test
Best
Mean Squared Error (mse)

10-10
10-2
Mu = 1e-08, at epoch 513
100

mu
10-4 10-5

10-10
10-6 Validation Checks = 0, at epoch 513
1

val fail
0
10-8

-1
0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500
513 Epochs 513 Epochs

(a) MSE 1 (b) TS 1

Best Validation Performance is 1.3786e-08 at epoch 311 Gradient = 4.5293e-06, at epoch 317
1010
Train
gradient

0 Validation
10
100
Test
Best
Mean Squared Error (mse)

10-10
10-2 Mu = 1e-09, at epoch 317
100
mu

10-5
10-4

10-10
Validation Checks = 6, at epoch 317
10
10-6
val fail

10-8 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
317 Epochs 317 Epochs

(c) MSE 2 (d) TS 2

Best Validation Performance is 6.2946e-08 at epoch 298 Gradient = 1.1099e-06, at epoch 304
1010
Train
gradient

0 Validation
10
100
Test
Best
Mean Squared Error (mse)

10-10
10-2 Mu = 1e-07, at epoch 304
100
mu

10-5
10-4

10-10
Validation Checks = 6, at epoch 304
10
10-6
val fail

-8
10 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
304 Epochs 304 Epochs

(e) MSE 3 (f) TS 3

Best Validation Performance is 2.4637e-07 at epoch 1000 Gradient = 1.7542e-05, at epoch 1000
1010
Train
gradient

0
10 Validation
100
Test
Best
Mean Squared Error (mse)

10-10
10-2
Mu = 1e-09, at epoch 1000
100
mu

10-4 10-5

10-10
10-6 Validation Checks = 0, at epoch 1000
4
val fail

10-8 10 2

0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
1000 Epochs 1000 Epochs

(g) MSE 4 (h) TS 4

Figure 4: MSE and TS measures for the FO-GUSM

 Error Histogram with 20 Bins
Function Fit for Output Element 1
3.5 800
Training
Training Targets Validation
Training Outputs 700
Test
3 Validation Targets
Validation Outputs Zero Error
Output and Target

Test Targets
600
Test Outputs
2.5

Instances
Errors 500
Fit

2 400

300
1.5

200
1
10-4 10 100
20 20 30 40 50 60 70 80 90 100

Targets - Outputs
0
Error

-0.00013
-0.00011
-8.5e-05
-6.2e-05

-1.8e-05
4.87e-06
2.72e-05
4.96e-05

9.44e-05
0.000117
0.000139
0.000162
0.000184
0.000206
0.000229
0.000251
0.000273
0.000296
-4e-05

7.2e-05
0

-2
Input Errors = Targets - Outputs

(a) Results 1 (b) EHs 1

Error Histogram with 20 Bins
Function Fit for Output Element 1
2.8 4000 Training
Training Targets Validation
2.6
Training Outputs 3500 Test
Validation Targets
2.4 Zero Error
Validation Outputs
Output and Target

Test Targets 3000

2.2
Test Outputs
Instances

Errors
2 2500
Fit
1.8
2000
1.6

1.4
1500

1.2 1000
1
10-3 10 500
50 20 30 40 50 60 70 80 90 100

Targets - Outputs
0 0
Error

0.00084

0.00864

0.01087
0.01198
-0.00919
-0.00807
-0.00696
-0.00585
-0.00473
-0.00362

-0.00139
-0.00027

0.001955
0.003069
0.004183
0.005297
0.006412
0.007526

0.009754
-0.0025

-5

-10
Input Errors = Targets - Outputs

(c) Results 2 (d) EHs 2

Error Histogram with 20 Bins
Function Fit for Output Element 1 4000
3 Training
Training Targets 3500 Validation
Training Outputs
Test
Validation Targets
2.5 Validation Outputs 3000 Zero Error
Output and Target

Test Targets
Test Outputs
2500
Instances

Errors
2 Fit
2000

1.5
1500

1000

1
0.01 0 10 20 30 40 50 60 70 80 90 100 500
Targets - Outputs
0 0
Error

0.01079

0.01501
0.0122
0.0136
-0.01169
-0.01029
-0.00888
-0.00747
-0.00607
-0.00466
-0.00326
-0.00185
-0.00045
0.000956
0.002361
0.003766
0.005171
0.006576
0.007981
0.009386

-0.01

-0.02
Input Errors = Targets - Outputs

(e) Results 3 (f) EHs 3

Error Histogram with 20 Bins
Function Fit for Output Element 1
3.5 4000 Training
Training Targets Validation
Training Outputs 3500 Test
3 Validation Targets
Validation Outputs Zero Error
Output and Target

Test Targets 3000

Test Outputs
2.5
Instances

Errors
2500
Fit

2 2000

1500
1.5

1000
1
10-3 10 500
50 20 30 40 50 60 70 80 90 100

Targets - Outputs
0 0
Error

11
0.00156

0.00389

0.00622
-0.00485
-0.00427
-0.00368

-0.00252
-0.00194
-0.00135
-0.00077
-0.00019
0.000395
0.000978

0.002143
0.002725
0.003308

0.004473
0.005055
0.005638
-0.0031

-5

-10
Input Errors = Targets - Outputs

(g) Results 4 (h) EHs 4

Figure 5: Assessment of error histograms and results for the FO-GUSM

 Training: R=1 Validation: R=1 Training: R=1 Validation: R=1
3 3
Output ~= 1*Target + -6.6e-09

Output ~= 1*Target + -1.3e-06

Output ~= 1*Target + 1.2e-06

Output ~= 1*Target + 1.4e-06

2.5 2.5
Data Data Data Data
Fit Fit Fit Fit
2.5 2.5
Y=T Y=T Y=T Y=T
2 2
2 2

1.5 1.5
1.5 1.5

1 1 1 1

0.5 0.5
0.5 0.5

0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
Target Target Target Target

Test: R=1 All: R=1 Test: R=1 All: R=1

3 3
Output ~= 1*Target + -9.7e-07

Output ~= 1*Target + -3.4e-08

Output ~= 1*Target + -7.1e-07

2.5 2.5
Data Data Output ~= 1*Target + 1e-06 Data Data
Fit Fit Fit Fit
2.5 2.5
Y=T Y=T Y=T Y=T
2 2
2 2

1.5 1.5
1.5 1.5

1 1 1 1

0.5 0.5
0.5 0.5

0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
Target Target Target Target

(a) Target 1 (b) Target 2

Training: R=1 Validation: R=1 Training: R=1 Validation: R=1
3 3
Output ~= 1*Target + -2.4e-05

Output ~= 1*Target + -1.4e-06

Output ~= 1*Target + 6.7e-08

Output ~= 1*Target + 2.4e-06

Data Data Data Data

2.5 2.5 Fit Fit
Fit 2.5 Fit 2.5
Y=T Y=T Y=T Y=T
2 2 2 2

1.5 1.5 1.5 1.5

1 1 1 1

0.5 0.5 0.5 0.5

0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
Target Target Target Target

Test: R=1 All: R=1 Test: R=1 All: R=1

3 3
Output ~= 1*Target + -2.7e-05

Output ~= 1*Target + -6.4e-06

Output ~= 1*Target + -2.4e-07

Output ~= 1*Target + -9.4e-07

Data Data Data Data

2.5 2.5 Fit Fit
Fit 2.5 Fit 2.5
Y=T Y=T Y=T Y=T
2 2 2 2

1.5 1.5 1.5 1.5

1 1 1 1

0.5 0.5 0.5 0.5

0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
Target Target Target Target

(c) Target 3 (d) Target 4

Figure 6: Regression for the FO-GUSM

12
 3.5 1.2
a=0.6 a=0.6
a=0.7 a=0.7
a=0.8 1 a=0.8
3 a=0.9 a=0.9

Occasional smoker L(t)

Potential smoker P(t)

0.8
2.5

0.6

2
0.4

1.5
0.2

1 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (day) Time (day)

(a) Assessment of the results for P(t) (b) Assessment of the results for L(t)
1.8 2.5
a=0.6 a=0.6
1.6 a=0.7 a=0.7
a=0.8 a=0.8
a=0.9 2 a=0.9
1.4

1.2
Chain smoker S(t)

Quit smoker Q(t)

1.5
1

0.8
1
0.6

0.4
0.5

0.2

0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (day) Time (day)

(c) Assessment of the results for S(t) (d) Assessment of the results for Q(t)

Figure 7: Evaluations of the outcomes using LMB-ANNs to address the FO-

GUSM

13
 -4 -4
a=0.6 a=0.6
-6 a=0.7 -6 a=0.7
a=0.8 a=0.8
a=0.9 a=0.9
-8 -8

Occasional smoker L(t)

Potential smoker P(t)

-10 -10

-12 -12

-14 -14

-16 -16

-18 -18

-20 -20
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (day) Time (day)

(a) AE for Potential smokers P(t) (b) AE for Occasional smokers L(t)
-4 -4
a=0.6 a=0.6
a=0.7 -6 a=0.7
-6
a=0.8 a=0.8
a=0.9 -8 a=0.9

-8
-10
Chain smoker S(t)

Quit smoker Q(t)

-10 -12

-12 -14

-16
-14
-18

-16
-20

-18 -22
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (day) Time (day)

(c) AE for Chain smoker S(t) (d) AE for Quit smoker Q(t)

Figure 8: Performance evaluations of the AE using LMB-ANNs to find the

solution of the FO-GUSM

14
samples can also be selected. If the values used for training are selected to be
> 75 percent, better performance can be attained due to the optimized input
bias values. The computational performances of LMB-ANNs were utilized to
generate numerical simulations for FO-GUSM, depicted in Figures 4, 5, and
6. Figure 4 specifically illustrates the MSE performances of FO-GUSM. The
best performance values achieved by FO-GUSM have been computed for FO
cases 1, 2, 3, and 4 are 2.1844e-09, 1.3786e-08, 6.2946e-08, and 2.4637e-07
with Epochs 513, 311, 298, and 1000. The gradient values have been com-
puted for 9.9378e-08, 4.5293e-06, 1.1099e-06, and 1.7542e-05 with Epochs
513, 317, 304, and 1000 for first, second, third, and fourth variation. The
results obtained and the performance of error histograms (EHs) for the FO-
GUSM using LMB-ANNs are depicted in Figure 5. The error histogram
metrics for the FO-GUSM have been calculated 4.87e-06, -0.00027, -0.00045,
and -0.00019. Figure 4 illustrates the optimal configurations for training,
testing, and validation of FO-MIM. Figure 6 displays the correlation values
obtained using LMB-ANNs for FO-GUSM through computational methods.
The accuracy of solving the fractional version of GUSM is detailed during
the validation, testing, and training phases. Figures 7 and 8 present compar-
isons of results and AE values for solving FO-GUSM. The evaluation of the
obtained solutions against the reference solutions is shown in Figure 7. The
effectiveness of the stochastic method is demonstrated by the alignment of
results. These exact matches confirm the validity and accuracy of the sug-
gested stochastic framework for addressing all types of FO-GUSM. Figure
8 displays the AE performances achieved using this proposed technique for
solving FO-GUSM. These AE values for the Potential smokers P (t) class are
depicted in Figure 8(a), and they have been computed as -9 to -19, -5 to
-16, -4 to -18, and -5 to -18. For the occasional smoker L(t) class, the AE
calculations have been conducted in the vicinity of -7 to -17, -6 to -18, -5 to
-14, and 6 to -17 in Figure 8(b). The AE values for the chain smoker S(t)
class are presented in Figure 8(c), achieved approximately -9 to -16, -6 to
-16, -5 to -15, and -6 to -16. The AE values for the final class quit smoker
Q(t), in Figure 8(d) have been computed as -8 to -16, -5 t -15, -4 to -15,
and -5 to -20. The computed AE metrics demonstrate the accuracy of the
stochastic LMB-ANNs procedure for FO-GUSM.

5 Conclusion
The objective of this article is to simulate numerically the fractional order
nonlinear mathematical model of smoking using computational methods re-
lying on SLMB-ANNs. Fractional order analyses have been introduced to

15
improve the accuracy of the GUSM compared to integer-order methods. The
solutions for three different types using fractional order derivatives have been
presented to demonstrate the numerical simulations for each classification of
the GUSM. Ten neurons were chosen along with a training set ratio of 75
percent, while the validation set accuracy was set at 10 percent, and the
testing set accuracy at 15 percent. To evaluate the effectiveness, accuracy,
capability, and suitability of the proposed computational framework, an ex-
tensive investigation was conducted. This investigation included correlation
analysis, MSE computation, EHs and STs information, as well as regres-
sion analysis. Additionally, AE calculations were performed effectively to
find the solution of the FO-GUSM. The effectiveness and importance of the
LMB-ANNs method were validated by achieving close agreement between
outcomes, accurate to 5 to 7 decimal places, in solving the GUSM. In fu-
ture studies, the suggested LMB-ANNs framework may be explored in areas
such as fluid dynamics, biological systems, wave propagation models, and
the non-linear dynamical systems [33, 34, 35, 36, 37].
Data Availability All data generated or analysed during this study are
included in this published article. Conflict of Interest There is no conflict
of interest in this research.

References
[1] A. Atangana, Mathematical model of survival of fractional calculus, crit-
ics and their impact: How singular is our world?, Advances in Difference
Equations 2021 (1) (2021) 403.

[2] A. Atangana, S. İğret Araz, Advanced analysis in epidemiological mod-

eling: Detection of wave, MedRxiv (2021) 2021–09.

[3] P. K. Roy, X. Cao, X.-Z. Li, P. Das, S. Deo, et al., Mathematical Analysis
and Applications in Modeling, Springer, 2020.

[4] M. Bachar, M. A. Khamsi, M. Bounkhel, A mathematical model for the

spread of covid-19 and control mechanisms in saudi arabia, Advances in
Difference Equations 2021 (1) (2021) 253.

[5] S. Tang, Z. Cai, Y. Fang, M. Lin, A new quantile treatment effect model
for studying smoking effect on birth weight during mother’s pregnancy,
Journal of Management Science and Engineering 6 (3) (2021) 336–343.

16
 [6] H. Singh, D. Baleanu, J. Singh, H. Dutta, Computational study of
fractional order smoking model, Chaos, Solitons & Fractals 142 (2021)
110440.
[7] A. Sharma, A. Misra, Backward bifurcation in a smoking cessation
model with media campaigns, Applied Mathematical Modelling 39 (3-4)
(2015) 1087–1098.
[8] G. ur Rahman, R. P. Agarwal, L. Liu, A. Khan, Threshold dynamics and
optimal control of an age-structured giving up smoking model, Nonlinear
Analysis: Real World Applications 43 (2018) 96–120.
[9] X. Li, R. P. Agarwal, J. Gomez-Aguilar, Q. Badshah, G. ur Rahman,
Threshold dynamics: Formulation, stability & sensitivity analysis of co-
abuse model of heroin and smoking, Chaos, Solitons & Fractals 161
(2022) 112373.
[10] I. Sofia, S. R. Bandekar, M. Ghosh, Mathematical modeling of smoking
dynamics in society with impact of media information and awareness,
Results in Control and Optimization 11 (2023) 100233.
[11] Z. Zhang, J. Zou, R. K. Upadhyay, Stability and hopf bifurcation of a
delayed giving up smoking model with harmonic mean type incidence
rate and relapse, Results in Physics 19 (2020) 103619.
[12] G. ur Rahman, R. P. Agarwal, Q. Din, Mathematical analysis of giv-
ing up smoking model via harmonic mean type incidence rate, Applied
mathematics and computation 354 (2019) 128–148.
[13] T. Saeed, Z. Sabir, M. S. Alhodaly, H. H. Alsulami, Y. G. Sánchez, An
advanced heuristic approach for a nonlinear mathematical based medical
smoking model, Results in Physics 32 (2022) 105137.
[14] H.-F. Huo, C.-C. Zhu, Influence of relapse in a giving up smoking model,
in: Abstract and Applied Analysis, Vol. 2013, Wiley Online Library,
2013, p. 525461.
[15] Q. Din, M. Ozair, T. Hussain, U. Saeed, Qualitative behavior of a smok-
ing model, Advances in Difference Equations 2016 (2016) 1–12.
[16] M. M. Rakočević, A harmonic structure of the genetic code, Journal of
theoretical biology 229 (2) (2004) 221–234.
[17] M. M. Rakocevic, Harmonic mean as a determinant of the genetic code,
arXiv preprint arXiv:1305.5103 (2013).

17
[18] M. Umar, Z. Sabir, M. A. Z. Raja, H. M. Baskonus, S.-W. Yao, E. Il-
han, A novel study of morlet neural networks to solve the nonlinear hiv
infection system of latently infected cells, Results in Physics 25 (2021)
104235.

[19] M. Umar, Z. Sabir, F. Amin, J. L. Guirao, M. A. Z. Raja, Stochastic

numerical technique for solving hiv infection model of cd4+ t cells, The
European Physical Journal Plus 135 (5) (2020) 403.

[20] D. Baleanu, F. A. Ghassabzade, J. J. Nieto, A. Jajarmi, On a new

and generalized fractional model for a real cholera outbreak, Alexandria
Engineering Journal 61 (11) (2022) 9175–9186.

[21] A. Jajarmi, D. Baleanu, K. Z. Vahid, H. M. Pirouz, J. Asad, A new and

general fractional lagrangian approach: a capacitor microphone case
study, Results in Physics 31 (2021) 104950.

[22] N. Ghassemi, A. Shoeibi, M. Rouhani, Deep neural network with gen-

erative adversarial networks pre-training for brain tumor classification
based on mr images, Biomedical Signal Processing and Control 57 (2020)
101678.

[23] T. Mahmood, N. Ali, N. I. Chaudhary, K. M. Cheema, A. H. Milyani,

M. A. Z. Raja, Novel adaptive bayesian regularization networks for peri-
staltic motion of a third-grade fluid in a planar channel, Mathematics
10 (3) (2022) 358.

[24] Z. Sabir, M. A. Z. Raja, M. Shoaib, J. G. Aguilar, Fmneics: fractional

meyer neuro-evolution-based intelligent computing solver for doubly sin-
gular multi-fractional order lane–emden system, Computational and Ap-
plied Mathematics 39 (4) (2020) 303.

[25] Z. Sabir, M. Umar, M. A. Z. Raja, D. Baleanu, Applications of gud-

ermannian neural network for solving the sitr fractal system, Fractals
29 (08) (2021) 2150250.

[26] Y. Guerrero Sánchez, Z. Sabir, H. Günerhan, H. M. Baskonus, Analyt-

ical and approximate solutions of a novel nervous stomach mathemat-
ical model, Discrete Dynamics in Nature and Society 2020 (1) (2020)
5063271.

[27] S. E. Awan, M. A. Z. Raja, A. Mehmood, S. A. Niazi, S. Siddiqa,

Numerical treatments to analyze the nonlinear radiative heat transfer in

18
 mhd nanofluid flow with solar energy, Arabian Journal for Science and
Engineering 45 (2020) 4975–4994.

[28] I. H. Qureshi, M. Awais, S. E. Awan, M. N. Abrar, M. A. Z. Raja,

S. O. Alharbi, I. Khan, Influence of radially magnetic field properties in
a peristaltic flow with internal heat generation: Numerical treatment,
Case Studies in Thermal Engineering 26 (2021) 101019.

[29] M. M. Khader, Using the generalized adams-bashforth-moulton method

for obtaining the numerical solution of some variable-order fractional
dynamical models, International Journal of Nonlinear Sciences and Nu-
merical Simulation 22 (1) (2021) 93–98.

[30] Z. Shah, M. Raja, Y.-M. Chu, W. Khan, S. Abbas, M. Shoaib, M. Ir-

fan, Computational intelligence of levenberg-marquardt backpropaga-
tion neural networks to study the dynamics of expanding/contracting
cylinder for cross magneto-nanofluid flow model, Physica Scripta 96 (5)
(2021) 055219.

[31] I. Khan, M. A. Z. Raja, M. Shoaib, P. Kumam, H. Alrabaiah, Z. Shah,

S. Islam, Design of neural network with levenberg-marquardt and
bayesian regularization backpropagation for solving pantograph delay
differential equations, IEEE Access 8 (2020) 137918–137933.

[32] S. Naz, M. A. Z. Raja, A. Kausar, A. Zameer, A. Mehmood, M. Shoaib,

Dynamics of nonlinear cantilever piezoelectric–mechanical system: An
intelligent computational approach, Mathematics and Computers in
Simulation 196 (2022) 88–113.

[33] H. M. Baskonus, H. Bulut, T. A. Sulaiman, New complex hyperbolic

structures to the lonngren-wave equation by using sine-gordon expansion
method, Applied Mathematics and Nonlinear Sciences 4 (1) (2019) 129–
138.

[34] N. Sweilam, S. Al-Mekhlafi, D. Baleanu, A hybrid stochastic fractional

order coronavirus (2019-ncov) mathematical model, Chaos, Solitons &
Fractals 145 (2021) 110762.

[35] T. Sajid, S. Tanveer, Z. Sabir, J. Guirao, Impact of activation energy

and temperature-dependent heat source/sink on maxwell–sutterby fluid,
Mathematical Problems in Engineering 2020 (1) (2020) 5251804.

19
[36] A. Yokuş, S. Gülbahar, Numerical solutions with linearization tech-
niques of the fractional harry dym equation, Applied Mathematics and
Nonlinear Sciences 4 (1) (2019) 35–42.

[37] E. Bonyah, M. Yavuz, D. Baleanu, S. Kumar, A robust study on the

listeriosis disease by adopting fractal-fractional operators, Alexandria
Engineering Journal 61 (3) (2022) 2016–2028.

20