Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology

ISSN No- 2584-2706

Comparative analysis of AGBFM and IWOFM

with Forecasting Models LSSVM-PSO,
LSSVM-ACO and LSSVM-WOA
Kuldeep Singh Malik
Tilak Raj Rohilla
Sandeep Kumar

Abstract: Performance analysis is utmost and optimizing network capacity extremely

important with the purpose of propelling the difficult (Teodorescu et al., 2023). Network
development of both software and hardware operators may effectively manage resources,
systems. By conducting a comprehensive avoid congestion, and improve end-user
study, it is possible to identify bottlenecks in service quality by using accurate prediction
the system and architecture, obtain essential models. In the context of Long-Term
information for selecting frameworks and Evolution (LTE), which supports
platforms, and ultimately result in contemporary mobile data communications,
improvements in performance. Performance traffic forecasting has emerged as a crucial
analysis is major step towards performance component of telecommunication network
optimizations. Various methods of management and optimization (Panjavarnam
optimization exist in order to achieve better et al., 2024). For efficient resource
performance of various machine learning allocation and to guarantee customer quality
algorithms. In this paper, performance of of service, it is crucial to comprehend traffic
proposed forecasting models is compared patterns and features, such as seasonality,
with various existing models. The trends, and the ongoing increase in mobile
performance of proposed models Improved cellular traffic. Traffic forecasting is useful
Whale Optimization Based Forecasting because it may give operators insight into
Model (IWOFM) and Adaptive Gradient future network demand, enabling them to
Based Forecasting Model (AGBFM) is prepare for and adjust to changes in traffic
compared after the final iteration of the patterns. Operators may make well-informed
forecasting models with LSSVM-PSO, judgments about network development,
LSSVM-ACO and LSSVM-WOA using the capacity improvements, and resource
similar dataset scenarios. allocation by examining historical data to
find trends and seasonal fluctuations
Introduction
(Ferreira et al., 2023b). This proactive
Predicting network traffic has been a crucial
approach mitigates network congestion,
issue in the telecom sector in recent years.
minimizes service disruptions, and enhances
The rapid rise in data services and mobile
the overall user experience.
device consumption has made monitoring

IJMSRT25APR007 www.ijmsrt.com 27
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

The proposed forecasting models (AGBFM Research Methodology

and IWOFM) are compared with three other The procedures used in this work's approach
models namely LSSVM-ACO, LSSVM- are depicted in Figure 1. Using historical
PSO and LSSVM-WOA (Sucarrat, 2021). data and an organized methodology, traffic
Despite the fact that meta-heuristic forecasting makes predictions about future
algorithms are highly effective in many traffic trends. Gathering pertinent historical
application areas, it has been noted that traffic data is the initial phase, after which it
LSSVM-ACO, LSSVM-PSO, and LSSVM- is pre-processed to guarantee its accuracy
WOA perform worse than the suggested and consistency. After that, a model is
ones when it comes to traffic burst chosen based on the data's properties. For
forecasting. This is due to thefact that the example, the AGBFM model could be the
first proposed model, The suggested best choice if the data shows significant
prediction model may be immediately seasonal trends (Hasan et al., 2024b). On the
integrated with the adaptive gradient-based other hand, IWOFM could be more
optimizer (AGBO) through parameter appropriate if the data has a complicated
optimization (gamma and structure with several variables. After the
sigma)(Chakrabarti & Chopra, 2023). The model has been chosen, the historical data is
second proposed model, IWOA enhances used to train it, and its parameters are
the explorationability by using inertia changed to best suit the data. Metrics like
weight factor whereas ACO and PSO cannot Mean Absolute Percentage Error (MAPE),
do optimizationand proper exploration using Root Mean Squared Error (RMSE), and
LSSVM(Suykens et al., 2002). Moreover, Coefficient of Determination (R2) are then
the search criteria of ACO and PSO include used to assess the model, which provide
exploitation instead of inherent gradient- insights into the model's performance. These
based optimization in AGBFM. Also, the indicators aid in determining which model is
usage of input data for tuning of parameters best suited for the job. The best model is
makes the proposed algorithms (AGBFM chosen to predict future traffic patterns after
and IWOFM) adaptive and efficient. The examination. To guarantee quality and
extra searches make the ACO and PSO consistency, the predicted values are then
loosely coupled optimizers (Manakkadu& post-processed. In order to make
Dutta, 2024). In case of PSO the comprehension and decision-making easier,
convergence did happen faster, butit did not the results are finally represented. The most
achieve good fitting cost. Compared to precise and trustworthy traffic predictions
ACO, the PSO achieved lower average are produced thanks to this methodical
fitting cost i.e. lower efficiencybut not better process, which supports well-informed
than AGBFM and IWOFM (Eberhart & Shi, judgments about infrastructure design and
2002).The evaluation parameters, namely, traffic management.
MSE, accuracy, TPR, FPR, precision andF1-
score are also calculated for all the existing
and proposed forecasting models.

IJMSRT25APR007 www.ijmsrt.com 28
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

START create raw data. Several commands might be

executed at the host computer's command
prompt to create network traffic. To
Description and Preparation
overload the network with traffic, the "ping
of the Dataset
command" and "apache server" 50
instructions are repeatedly executed in this
Model for traffic case. The host computer uses Wire shark to
Forecasting record these data bursts. Live network data
is concurrently captured by the Wire shark
capture engine from many network
Performance Evaluation interfaces. The subset of captured traffic is
Metrics for Traffic Forecasting
shown in figure 2.
END
Figure 2. The Wireshark engine is
operating to capture network traffic
Figure 1. An illustration of the research When capturing of data gets completed for a
approach flow certain time interval, it is saved in .csv form.
In the present study, firstly, normal data is
3.1. Dataset Description and Preparation captured without any traffic burst and
plotted in a graph and then VMs (virtual
The data which is used for network traffic machines) are flooded with data packets by
forecasting is taken using two different running commands on prompt very
scenarios: Scenario I: Network traffic is frequently and these packets are captured.
flooded into several virtual computers to The normal and exponential bursts seen in

IJMSRT25APR007 www.ijmsrt.com 29
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

captured traffic is shown in figure 3 and 4

respectively. Figure 4. Exponential (bursty) network
traffic laid over normal traffic data captured
through Wire shark
Scenario II: In this case, the data used to
evaluate the present study is taken from
traffic (GB/micro-sec)

CRAWDAD (Community Resource for

Archiving Wireless Data) iitkdg/app traffic
datasets of a Smartphone app collected using
tcpdump. The computer program Tcpdump
time(micr is a command-line interface-based data-
network packet analyzer. Through a network
that the system is linked to, it enables the
user to view TCP/IP and other packets that
are transmitted and received. The desired
traffic for evaluation came from Google
Hangout of smartphone app. An application
called google hangout [12] traffic
(GB/micro-sec) time(micro-seconds) traffic
(GB/micro-sec) time(micro-seconds) 52
facilitates its users to do chats, carry out
VoIP calls and video calls. Google hangout
is not a completely peer-to-peer service
platform, although it has features of a peer-
Figure3.Normal network traffic topeer application as it permits two users to
bursts data captured through interconnect in real-time using a session
Wireshark server which is selected dynamically. In the
form of .pcap files, data was collected with
only the Google Hangouts app running in
the foreground and only required system
traffic(GB/micro-sec)

functions running in the background. A

portion of data from the CRAWDAD
community's Google Hangouts smartphone
app is shown in table 3.1.

Figure 5. Exponential (burst) network

traffic overlaid over normal traffic data

time(micro

IJMSRT25APR007 www.ijmsrt.com 30
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

3.2. Models for Traffic Forecasting ii. Safe Zone: Network administrators can
In 5G base stations, traffic forecasting is successfully employ the safe zone—
essential for effective resource allocation, identified by the forecasting model—as
network management, and quality of service an alarm system. It may be described as
assurance. Traditional statistical models like the area of network traffic that is within
ARIMA are among the models that have typical bounds. In the safe zone, there are
been put up and evaluated for this purpose no traffic spikes that might cause
(Rizkya et al., 2019), gradient-based network-wide congestion or other
machine learning models, such as AGBFM associated issues. In order to take the
(Amara-Ouali et al., 2022), and techniques proper steps to address this issue, it is
like IWOFM. This section provides a utilized to pinpoint the precise moment
theoretical framework for traffic forecasting the network peaked.
with these models, highlighting their iii. Mean Square Error (MSE): In
advantages and potential uses (Ruan et al., forecasting, MSE is a frequently used
2016). performance metric. The mean square
3.3. Traffic Forecasting Performance error (MSE) is one of the main metrics
Evaluation Metrics used to assess predicting effectiveness
The performance analysis in this study takes (Thompson, 1990). The average of the
into account a number of parameters that sum of squares of forecast errors is used
have been proposed in the literature. to compute the MSE (Ahmar, 2020).
Selecting the right parameter is crucial for iv. Root Mean Square Error (RMSE): It is
comparing the suggested model to the ones the predicting mistakes' standard
that already exist. Forecasting accuracy may deviation. The distance between the data
be expressed using a variety of criteria. A points and the regression line is measured
brief overview of the parameters explained by forecasting errors. It indicates the
by authors in their papers is listed below: degree of data concentration around the
i. Confidence Intervals: It is considered as line of best fit. The square root of MSE is
the most widely used parameter for the mathematical definition of RMSE
network traffic forecasting. A confidence (Kumar et al., 2021b).
interval quantifies the uncertainty on an v. Confusion Matrix: The forecasting
estimated traffic, such as the mean or model's prediction accuracy is expressed
standard deviation (Salas et al., 2003). using this statistic. A confusion matrix,
Confidence intervals are most commonly where N is the number of target classes,
used when forecasting with a regression
model, where a quantity is being
predicted. Confidence intervals tell how
well the model has determined a
parameter of interest, such as a mean or
regression coefficient (Bartkiewicz,
2000).

IJMSRT25APR007 www.ijmsrt.com 31
 Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

is a 𝑁 × 𝑁 matrix used to assess aviii. Accuracy: It is the main metric used to

forecasting model's performance (Patro & compare the performance of various models.
Patra, 2014). The forecasting model's Its definition is the proportion of cases that
anticipated values and the actual target were accurately anticipated.
values are compared in the matrix. Thisix. Precision: The percentage of true positives
provides a comprehensive picture of the reported by the suggested model relative to
model's performance and the kind of all positives is known as precision.
mistakes it is making. The confusion x. F1-Score:The F1 Score is the weighted
matrix, which includes the number of true average of Precision and Recall. Therefore,
positives, true negatives, false positives, both false positives and false negatives are
and false negatives, is displayed in Figure taken into account in this score. F1 is more
6. The following equations can be used to helpful, particularly in situations when the
assess a forecasting model's performance distribution of classes is uneven. Accuracy
in terms of sensitivity or true positive performs best when the costs of false
rate, specificity, accuracy, and precision positives and false negatives are equal.
xi. Execution Time:CPU cycles are used to
Figure 6. Confusion Matrix. compute the execution time. It is equivalent
to the number of CPU cycles required to run
True Positive (TP) indicates that both the the model with a size n dataset.
model's predicted and actual values were xii. Computational Complexity:It has a direct
positive. correlation with the quantity of resources
True Negative (TN) indicates that the needed to execute an algorithm.
model anticipated a negative value, while
the actual result was negative. 4. Result and Discussion
False Positive (FP) occurs when the model 4.1 Total Traffic Forecast
predicted a positive result but the actual
value was negative. Several metrics are used to compare the
False Negative (FN) occurs when the model predicting outcomes produced by the
predicted a negative result but the actual suggested and current models during the
value was positive. testing period with the actual total traffic
vi. True Positive Rate (TPR) or Sensitivity or data. The y-axis shows the amount of traffic,
Recall: This is an additional metric used to while the x-axis shows the time in hours. To
assess a forecasting model's performance. It see how well each model performs, the
is calculated by dividing the total number of prediction is placed next to the real traffic
FP by the total number of TP. The data. The graph gives a clear visual
likelihood that a real positive will test depiction of each model's accuracy by
positive is known as TPR. highlighting differences between expected
vii. False Positive Rate: It is calculated by and actual values. Algorithms for training
dividing the total number of incorrect and optimization make up forecasting
positive forecasts by the total number of models.ACO (Ant Colony Optimization),
negative forecasts. PSO (Particle Swarm Optimization), WOA

IJMSRT25APR007 www.ijmsrt.com 32
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

(Whale Optimization Algorithm), IWOA

(Improved Whale Optimization), and AGO
(Adaptive Gradient Optimization) are
utilized as optimization algorithms, while
LSSVM (Least Square Support Vector
Machine) is employed for training. Figure 7
displays the convergence
ence curve for both the
suggested and current models.

Iteration (#)

Figure 8. LSSVM-PSO
PSO Convergence
Curve

Figures 7 and 8 display the optimal

optimization cost that may be achieved with
the LSSVM classifier er utilizing ACO and
PSO. Even after 50 iterations, ACO Pareto
optimization was not obtained. As a result,
LSSVM-ACOACO has higher optimization
efficiency. This is a result of ACO's slower
search parameters. Figures 9, 10, 11, 12, and
13 display the confusion matrix for LSSVM-
LSSVM
Figure7. LSSVM-ACO
ACO Convergence
ACO, LSSVM-PSO,
PSO, LSSVM-WOA,
LSSVM
Curve
AGBFM, and IWOFM for scenarios 1 and
2, respectively.

IJMSRT25APR007 www.ijmsrt.com 33
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

Confusion matrix in scenario1

Figure 9: Confusion Matrix of LSSVM-ACO

Confusion matrix for scenario1

Figure 10: Confusion Matrix of LSSVM-PSO

IJMSRT25APR007 www.ijmsrt.com 34
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

Confusion matrix forscenario1

Figure 11: Confusion Matrix of LSSVM-WOA

Figure 12: Confusion Matrix of AGBFM

IJMSRT25APR007 www.ijmsrt.com 35
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

Figure 13: Confusion Matrix of IWOFM

4.2 Performance Analysis AGBFM. Also, the usage of input data for
The proposed forecasting models (AGBFM tuning of parameters makes the proposed
and IWOFM) are compared with three other algorithms (AGBFM and IWOFM) adaptive
models namely LSSVM-ACO, LSSVM- and efficient.
PSO and LSSVM-WOA. Despite the fact The ACO and PSO optimizers are loosely
that meta-heuristic algorithms are highly connected due to the additional searches.
effective in many application areas, it has Although convergence occurred more
been noted that LSSVM-ACO, LSSVM- quickly in the PSO instance, an acceptable
PSO, and LSSVM-WOA perform worse fitting cost was not attained. The PSO
than the suggested ones when it comes to obtained lower average fitting cost (i.e.,
traffic burst forecasting.This is because lesser efficiency) than ACO, but it was not
parameter optimization (gamma and sigma) superior than AGBFM and IWOFM. For
may be used to directly integrate the first each of the current and suggested
suggested model, the adaptive gradient forecasting models, the evaluation
based optimizer (AGBO), with the parameters—MSE, accuracy, TPR, FPR,
suggested prediction model. The second precision, and F1-score—are also computed.
proposed model, EWOA enhances the
exploration ability by using inertia weight 4.2.1 Mean Square Error
factor whereas ACO and PSO cannot do Both the suggested and current algorithms'
optimization and proper exploration using MSEs are assessed. Figure 14 displays the
LSSVM. Moreover, the search criteriaof outcomes for scenarios 1 and 2. MSE is a
ACO and PSO include exploitation instead method for assessing how well forecasts or
of inherent gradient-based optimization in estimates match actual values. This is used
as a model assessment metric for regression

IJMSRT25APR007 www.ijmsrt.com 36
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

models, where a lower value denotes a better other models, the IWOA optimizer may be
fit. Because the improved whale closely integrated with the LSSVM training
optimization model is used to optimize the method, which makes this model suitable for
training model's hyperparameters, the forecasting.
IWOFM has the lowest MSE. Compared to

MSE of Different
MSE of Different
Forecasting Models
Forecasting Models
29.20%
27.90%

8.10%
7.40% 4.70%
3.42% 3.30%
4.20% 2.94%
3.16%

MSE FOR SCENARIO2

MSE FOR SCENARIO1
LSSVM-ACO LSSVM-PSO LSSVM-WOA
LSSVM-ACO LSSVM-PSO LSSVM-WOA
AGBFM IWOPM
AGBFM IWOPM

Figure14. MSE of forecasting models in scenario 1 and scenario 2

4.2.2. Accuracy improves their ability to explore the search
The accuracy is tested for the proposed and space and, as a consequence, raises the
current algorithms. Figure 15 displays the overall accuracy of forecasting findings,
outcomes for both scenarios 1 and 2. The which is why the IWOFM exhibits
three-dimensional search criteria in this maximum accuracy.
model, which search agents utilize,

IJMSRT25APR007 www.ijmsrt.com 37
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

Accuracy of Different Accuracy of Different

Forecasting Models Forecasting Models

96.60% 95.40%
92.60% 91.80% 94.10% 90.70% 93.40%
90.80%

72.20% 70.60%

ACCURACY FOR SCENARIO 1 ACCURACY FOR SCENARIO 2

LSSVM-ACO LSSVM-PSO LSSVM

LSSVM-WOA LSSVM-ACO LSSVM-PSO LSSVM-WOA

AGBFM EWOFM AGBFM EWOFM

Figure15. Accuracy of forecasting models in both the suggested and current methods.
scenario 1 and scenario 2 Because there are more genuine alarms as
the number of iterations rises, the TPR
4.2.3. True Positive Rate (TPR) or reaches
ches its maximum in the case of IWOFM.
Sensitivity or Recall and False Positive Confirmation is shown by the increased
Rate (FPR) TPR. In the case of IWOFM,
IWOFM the FPR is
Figure
re 16 displays the genuine positive rate lowest. A lower FPR rating is preferable as
for both the suggested and current methods. it indicates inaccurate predictions.
Figure 17 displays the false positive rate for

IJMSRT25APR007 www.ijmsrt.com 38
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

TPR of Different TPR of Different

Forecasting Models Forecasting Models
0.984285
0.988553 0.970347
0.958315 0.957921
0.979391
0.966805 0.963018

0.848367
0.858614

TPR(SENSITIVITY) FOR SCENARIO 2

TPR(SENSITIVITY) FOR SCENARIO 1
LSSVM-ACO LSSVM-PSO LSSVM-WOA LSSVM-ACO LSSVM-PSO
PSO LSSVM-WOA
AGBFM EWOFM AGBFM EWOFM

Figure 16. TPR of forecasting models in scenario 1 and scenario 2

FPR of Different FPR of Different

Forecasting Models Forecasting Models
0.47264
0.493136

0.172813
0.15738 0.159227 0.191427 0.192952
0.09438 0.14237
0.125227

FPR FOR SCENARIO 1 FPR FOR SCENARIO 2

LSSVM-ACO LSSVM-PSO LSSVM-WOA LSSVM-ACO LSSVM-PSO

PSO LSSVM-WOA
AGBFM EWOFM AGBFM EWOFM

marginally higher than those of the other

Figure 17. FPR of forecasting models in scenario 1 and scenario
models. The2total value is superior
supe than the
other models since the F1-score
score provides the
4.2.4. F1-Score harmonic mean of these two. Figure 18
Since the IWOFM model has the best TPR displays the F1-Score
Score for the suggested and
(or recall) and accuracy, its F1-score
score is current methods.

IJMSRT25APR007 www.ijmsrt.com 39
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

F1-Score
Score of Different F1-Score
Score of
Forecasting Models Different
Forecasting Models
0.977365
0.9459 0.939873 0.959615
0.969137
0.932413 0.931791 0.951603
0.784678
0.770488

F1-SCORE
SCORE FOR SCENARIO 1 F1-SCORE
SCORE FOR SCENARIO 2

LSSVM-ACO LSSVM-PSO LSSVM-WOA LSSVM-ACO LSSVM-PSO LSSVM-WOA

AGBFM EWOFM AGBFM EWOFM

Figure 18. F1-Score

Score of forecasting models in LSSVM-WOA WOA for both the scenarios of
scenario 1 and scenario 2 dataset is summarized in table 1 and table 2.
Table 1. Comparative analysis of existing
The performance evaluation
luation of IWOFM, and proposed forecasting models for
AGBFM, LSSVM-PSO,
PSO, LSSVM
LSSVM-ACO and scenario 1
Parameters
LSSVM-- LSSVM- LSSVM- AGBFM IWOFM
ACO PSO WOA
Forecasting
Models
Mean Square Error 7.4% 27.9% 4.2% 3.16% 2.94%
TP alarms 21698 16931 21509 23619 24267
FP alarms 1737 6504 1926 1491 843
TN alarms 9300 7257 9219 7873 8089
FN alarms 745 2788 826 497 281
Accuracy 92.60% 72.20% 91.80% 94.10% 96.60%
TPR (Sensitivity) 0.966805 0.858614 0.963018 0.979391 0.988553
FPR 0.15738 0.47264 0.172813 0.159227 0.09438
Precision 0.92588 0.722466 0.917815 0.940621 0.966428
F1-Score 0.9459 0.784678 0.939873 0.959615 0.977365
Execution Time* 56 81 49 45 41

IJMSRT25APR007 www.ijmsrt.com 40
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

Computational 𝜃(𝑛 log 𝑛) + 𝜃(𝑛 log 𝑛) 𝜃(𝑛𝑙𝑜𝑔 𝑛) 𝜃(𝑛𝑙𝑜𝑔 𝑛) 𝜃(𝑛𝑙𝑜𝑔 𝑛)

Complexity ** + 𝜃(𝑛 ) + 𝜃(𝑛 ) + 𝜃(𝑛 )

Table 2. Comparative analysis of existing and proposed forecasting models for scenario 2

Parameters
LSSVM- LSSVM- LSSVM- AGBFM IWOFM
Forecasting ACO PSO WOA
Models

Mean Square Error 8.1% 29.2% 4.7% 3.42% 3.30%

TP alarms 15403 11973 15389 15838 17349
FP alarms 1563 4993 1577 1127 828
TN alarms 6602 5132 6596 6789 5784
FN alarms 670 2140 676 484 277
Accuracy 90.80% 70.60% 90.70% 93.40% 95.40%
TPR (Sensitivity) 0.958315 0.848367 0.957921 0.970347 0.984285
FPR 0.191427 0.493136 0.192952 0.14237 0.125227
Precision 0.907875 0.705706 0.907049 0.933569 0.954448
F1-Score 0.932413 0.770488 0.931791 0.951603 0.969137
Execution Time* 60 85 53 51 46
Computational 𝜃(𝑛 log 𝑛) + 𝜃(𝑛 log 𝑛) 𝜃(𝑛𝑙𝑜𝑔 𝑛) 𝜃(𝑛𝑙𝑜𝑔 𝑛) 𝜃(𝑛𝑙𝑜𝑔 𝑛)
Complexity ** + 𝜃(𝑛 ) + 𝜃(𝑛 ) + 𝜃(𝑛 )

3. The two proposed models are compared

with the existing models to show the
5. Conclusion and Future Scope effectiveness of the proposed ones using
1. A study of the various machine learning the performances matrices. The proposed
algorithms is performed in order to forecasting models are compared with
choose the suitable algorithm for training. three existing models namely LSSVM-
LSSVM machine learning algorithm is ACO, LSSVM-PSO and LSSVM-WOA.
considered for training using dataset. While meta-heuristic algorithms are
Various optimization schemes are also effective in many areas of application, it
studied to optimize the parameters of has been found that both LSSVM-ACO
machine learning algorithm for better and LSSVM-PSO perform worse than
performance. deterministic AGBFM in traffic burst
2. The proposed models have been prediction. IWOFM outperforms all
compared on the evaluation parameters models in forecasting because it
using confusion matrix namely MSE, concentrates on finding the near-optimal
TPR, FPR, accuracy, execution time and solution to the problem, and whale-based
computational complexity precision, f1- optimizers offer a high-level framework.
score. 4. In future, IWOFM and AGBFM models
can also be extended further, so as to

IJMSRT25APR007 www.ijmsrt.com 41
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

apply to different real world problems optimal group formation based on previous
like image processing, data mining and academic performance. Machine Learning
feature selection. Research, 9(2), 48–52.
https://doi.org/10.11648/j.mlr.20240902.13
References
Ahmar, A. S. (2020). Forecast Error Hasan, M., Babu, M. S., & Emran, M. A.
Calculation with Mean Squared Error (2024b). A machine learning approach to
(MSE) and Mean Absolute Percentage Error optimal group formation based on previous
(MAPE). JINAV Journal of Information and academic performance. Machine Learning
Visualization, 1(2), 94–96. Research, 9(2), 48–52.
https://doi.org/10.35877/454ri.jinav303 https://doi.org/10.11648/j.mlr.20240902.13
Amara-Ouali, Y., Fasiolo, M., Goude, Y., &
Yan, H. (2022). Daily peak electrical load Kumar, A., Alsadoon, A., Prasad, P. W. C.,
forecasting with a multi-resolution Abdullah, S., Rashid, T. A., Pham, D. T. H.,
approach. International Journal of & Nguyen, T. Q. V. (2021a). Generative
Forecasting, 39(3), 1272–1286. adversarial network (GAN) and Improved
https://doi.org/10.1016/j.ijforecast.2022.06.0 root mean square error (ERMSE): deep
01 learning for stock price movement
prediction. Multimedia Tools and
Bartkiewicz, W. (2000). Confidence Applications, 81(3), 3995–4013.
intervals prediction for the short-term https://doi.org/10.1007/s11042-021-11670-
electrical load neural forecasting models. w
E+I Elektrotechnik Und
Informationstechnik, 117(1), 8–12. Lira, J. M. S., Salgado, E. G., &Beijo, L. A.
https://doi.org/10.1007/bf03161392 (2019). ARIMA models as an alternative to
predict the diffusion of the ISO 14001
Chakrabarti, K., & Chopra, N. (2023). A standard in Europe. Journal of
control theoretic framework for adaptive Environmental Planning and Management,
gradient optimizers. Automatica, 160, 63(2), 275–286.
111466. https://doi.org/10.1080/09640568.2019.1577
https://doi.org/10.1016/j.automatica.2023.11 721
1466
Manakkadu, S., & Dutta, S. (2024). Ant
Ferreira, G. O., Ravazzi, C., Dabbene, F., Colony Optimization based Support Vector
Calafiore, G. C., & Fiore, M. (2023). Machine for Improved Classification of
Forecasting Network Traffic: A survey and Unbalanced Datasets. Procedia Computer
tutorial with Open-Source Comparative Science, 237, 586–593.
Evaluation. IEEE Access, 11, 6018–6044. https://doi.org/10.1016/j.procs.2024.05.143
https://doi.org/10.1109/access.2023.3236261
Panjavarnam, B., Poornima, P., & Subathra,
Hasan, M., Babu, M. S., & Emran, M. A. K. (2024). Network traffic prediction in 4G-
(2024a). A machine learning approach to LTE using machine learning techniques.

IJMSRT25APR007 www.ijmsrt.com 42
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

2022 International Conference on computer science (pp. 441–448).

Communication, Computing and Internet of https://doi.org/10.1007/3-540-44869-1_56
Things (IC3IoT), 13, 1–6.
https://doi.org/10.1109/ic3iot60841.2024.10 Sucarrat, G. (2021). Identification of
550364 volatility proxies as expectations of squared
financial returns. International Journal of
Patro, N. M., & Patra, M. R. (2015). A Forecasting, 37(4), 1677–1690.
Novel Approach to Compute Confusion https://doi.org/10.1016/j.ijforecast.2021.03.0
Matrix for Classification of n-Class 08
Attributes with Feature Selection.
Transactions on Machine Learning and Suykens, J. a. K., Van Gestel, T., De
Artificial Intelligence. Brabanter, J., De Moor, B., & Vandewalle,
https://doi.org/10.14738/tmlai.32.1108 J. (2002). Least squares support vector
machines.
Patro, V. M., & Patra, M. R. (2014). https://doi.org/10.1142/9789812776655
Augmenting Weighted Average with
Confusion Matrix to Enhance Classification Teodorescu, C. A., Durnoi, A. C., & Vargas,
Accuracy. Transactions on Machine V. M. (2023). The rise of the mobile
Learning and Artificial Intelligence, 2(4). Internet: Tracing the evolution of portable
https://doi.org/10.14738/tmlai.24.328 devices. Proceedings of the . . . International
Conference on Business Excellence, 17(1),
Rizkya, I., Syahputri, K., Sari, R. M., 1645–1654. https://doi.org/10.2478/picbe-
Siregar, I., &Utaminingrum, J. (2019). 2023-0147
Autoregressive Integrated Moving Average
(ARIMA) model of forecast demand in Thompson, P. A. (1990). An MSE statistic
distribution centre. IOP Conference Series for comparing forecast accuracy across
Materials Science and Engineering, 598(1), series. International Journal of Forecasting,
012071. https://doi.org/10.1088/1757- 6(2), 219–227. https://doi.org/10.1016/0169-
899x/598/1/012071 2070 (90)90007-x

Ruan, W., Sheng, Q. Z., Xu, P., Tran, N. K.,

Falkner, N. J., Li, X., & Zhang, W. E. (2016,
October 24). Forecasting Seasonal Time
Series Using Weighted Gradient RBF
Network based Autoregressive Model.
https://doi.org/10.1145/2983323.2983899

Salas, R., Torres, R., Allende, H., &

Moraga, C. (2003). Robust Estimation of
Confidence Interval in Neural Networks
applied to Time Series. In Lecture notes in

IJMSRT25APR007 www.ijmsrt.com 43