Impact of Signal Segmentation on EEG-Based Seizure Detection: A

Comparative Time-Frequency Analysis

Nuri Ikizler

and Gunes Ekim

Department of Electronics and Automation, Trabzon Vocational School, Karadeniz Technical University, Trabzon, Turkey

Keywords: EEG Signal Segmentation, Epileptic Seizure Detection, Random Forest, Discrete Wavelet Transform, Power

Spectral Density.

Abstract: Accurate and timely detection of epileptic seizures from EEG signals is essential for reliable clinical decision

support and patient monitoring. In this study, the impact of data segmentation on seizure detection

performance is systematically investigated using the publicly available EEG dataset from the University of

Bonn. Two commonly applied feature extraction methods, Discrete Wavelet Transform and Power Spectral

Density, are evaluated in combination with a Random Forest classifier across multiple segmentation levels.

A fully automated experimental framework is developed in MATLAB, and classification tasks of varying

complexity, including binary and multi-class problems, are considered. The results reveal that signal

segmentation significantly affects classification performance, with moderate segmentation generally

improving accuracy for both Discrete Wavelet Transform and Power Spectral Density features. While

excessive segmentation degrades performance in the Discrete Wavelet Transform based approach, the Power

Spectral Density based method demonstrates greater robustness across segmentation levels. These findings

underline the critical role of segmentation strategy in EEG-based seizure detection and highlight the

importance of optimizing this parameter based on the chosen feature extraction technique. The insights

obtained from this study can guide the development of more efficient, real-time, and clinically applicable

seizure monitoring systems.

1 INTRODUCTION

Epilepsy is a chronic neurological disorder

characterized by recurrent, unprovoked seizures

resulting from abnormal electrical activity in the

brain. Affecting over 50 million individuals

worldwide, epilepsy significantly impairs quality of

life and, in severe cases, poses life-threatening risks

(World Health Organization, 2025). Accurate

detection and monitoring of epileptic seizures are

essential for effective disease management, yet

conventional diagnosis heavily relies on manual

inspection of electroencephalogram (EEG)

recordings by trained specialists. This process is time-

consuming, labour-intensive, and prone to subjective

interpretation, especially in long-term monitoring

scenarios (Milligan, 2021).

To address these challenges, automated seizure

detection systems based on EEG signal analysis have

https://orcid.org/0000-0002-7632-1973

https://orcid.org/0000-0003-4867-3100

been extensively investigated in recent years (Naidu

and Zuva, 2023). Numerous studies have explored

different approaches for extracting discriminative

features from EEG recordings, ranging from time-

domain methods to advanced time-frequency and

spectral techniques. Among these, Discrete Wavelet

Transform (DWT) and Power Spectral Density (PSD)

have gained significant attention due to their ability

to capture both transient and stationary characteristics

of EEG signals associated with seizure activity (Liu

et al., 2023, Kinaci et al., 2024).

In parallel, various machine learning algorithms,

including Support Vector Machines (SVM), k-

Nearest Neighbors (k-NN), and ensemble classifiers

such as Random Forests, have been employed to

classify extracted features with promising results

(Siddiqui et al., 2021). Despite these advancements,

many existing studies rely on pre-segmented datasets

or fixed-length signals, often overlooking the critical

Ikizler, N. and Ekim, G.

Impact of Signal Segmentation on EEG-Based Seizure Detection: A Comparative Time-Frequency Analysis.

DOI: 10.5220/0014284900004848

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences (ICEEECS 2025), pages 305-312

ISBN: 978-989-758-783-2

305

role of signal segmentation strategy in the overall

classification performance (Thangavel, 2022).

Moreover, the selection and configuration of

segmentation parameters remain largely empirical in

literature, and their interaction with feature extraction

techniques is not systematically explored. This gap is

particularly relevant for real-world clinical

applications, where signal length, processing time,

and system responsiveness are key considerations.

The aim of this study is to systematically

investigate how different signal segmentation

strategies affect seizure detection performance using

two widely adopted feature extraction methods, DWT

and PSD, in combination with Random Forest

classification. To ensure comprehensive evaluation,

experiments were conducted on the well-established

University of Bonn EEG dataset, which is frequently

utilized as a benchmark in the field.

Unlike many previous studies, this work focuses

specifically on the relationship between segmentation

granularity, feature extraction approach, and

classification accuracy. The results demonstrate that

appropriate segmentation can significantly enhance

detection performance, while suboptimal

segmentation may degrade system reliability. These

findings not only contribute to a better understanding

of the signal processing pipeline for EEG-based

seizure detection but also provide practical insights

for developing more robust, real-time, and clinically

applicable monitoring systems.

2 MATERIAL AND METHODS

Block diagram of proposed study is given in Figure 1.

2.1 EEG Dataset

This study utilizes the publicly available EEG dataset

provided by the Department of Epileptology at the

University of Bonn, which has been extensively used

for seizure detection research. The dataset consists of

five subsets, each containing 100 single-channel EEG

recordings. Sets A and B represent surface EEG

recordings from healthy individuals with eyes open

and eyes closed, respectively. Sets C, D, and E

contain intracranial EEG recordings from epilepsy

patients, with set E specifically representing seizure

activity (Andrzejak et al., 2001).

Each EEG recording is composed of 4096

samples, acquired at a sampling frequency of 173.61

Hz. To investigate the effect of signal segmentation

on classification performance, the recordings were

divided into smaller, equally sized segments.

Different segmentation scenarios were applied,

including 1 (no segmentation), 2, 4, 8, and 16

segments per signal, allowing for a systematic

evaluation of how segment length influences feature

extraction and subsequent classification.

Segmenting the signals into smaller portions

provides both an increased number of training

examples and an opportunity to capture localized

signal variations, which is particularly relevant for the

detection of transient events such as epileptic

seizures.

2.2 Data Segmentation

Signal segmentation was performed by evenly

dividing each 4096-sample EEG recording into

smaller non-overlapping segments based on the

chosen segmentation factor. For instance, applying a

segmentation factor of 2 results in segments of 2048

samples each, whereas a factor of 16 yields segments

of 256 samples.

Figure 1: Block diagram of proposed study.

This segmentation process serves multiple

purposes. Firstly, it increases the total number of

available data samples, which is beneficial for

ICEEECS 2025 - International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences

306

training machine learning models and reducing the

risk of overfitting. Secondly, it allows for finer

temporal analysis by focusing on shorter signal

windows, which can reveal localized patterns and

frequency components that may be less visible in

longer segments. Importantly, the segmentation

strategy can influence the ability of feature extraction

methods to capture relevant information, making its

optimization a critical step in EEG-based

classification tasks (Zhou et al, 2024).

2.3 Feature Extraction

Two distinct feature extraction techniques were

applied to characterize the EEG signal segments:

Discrete Wavelet Transform (DWT) and Power

Spectral Density (PSD) analysis. Both approaches

aim to quantify the essential temporal and spectral

properties of the EEG signals by generating 10-

dimensional feature vectors for each segment.

2.3.1 Discrete Wavelet Transform Features

Discrete Wavelet Transform (DWT) provides a

multiresolution time-frequency analysis of the EEG

signals, effectively capturing both low and high

frequency components (Almahdi et al., 2021, Subekti

et al., 2024). In this study, each EEG segment was

decomposed into four levels using the Daubechies 4

(db4) mother wavelet. From the resulting

approximation and detail coefficients, the following

10 statistical features were extracted to form the

DWT-based feature vector:

 D

: Mean of the approximation coefficients at

level 4.

 D

: Standard deviation of the approximation

coefficients at level 4.

 D

: Mean of the detail coefficients at level 4.

 D

: Standard deviation of the detail coefficients

at level 4.

 D

: Mean of the detail coefficients at level 3.

 D

: Standard deviation of the detail coefficients

at level 3.

 D

: Mean of the detail coefficients at level 2.

 D

: Standard deviation of the detail coefficients

at level 2.

 D

: Mean of the detail coefficients at level 1.

 D

: Standard deviation of the detail

coefficients at level 1.

These features collectively capture signal energy

distribution, variability, and frequency content across

multiple scales, which are essential for distinguishing

seizure activity from normal brain signals.

2.3.2 Power Spectral Density Features

To characterize the frequency-domain properties of

the EEG segments, the Welch method was applied to

estimate the PSD of each segment and its

corresponding reference signal. Based on the PSD

distributions, the following 10 features were

calculated to construct the PSD-based feature vector:

 P

: Kullback-Leibler divergence between the

segment PSD and the reference PSD.

 P

: L2-norm (Euclidean distance) between the

segment PSD and the reference PSD.

 P

: Difference in total spectral power between

the segment and reference PSD.

 P

: Difference in spectral entropy between the

segment and reference PSD.

 P

: Difference in spectral flatness between the

segment and reference PSD.

 P

: Difference in spectral bandwidth between

the segment and reference PSD.

 P

: Frequency corresponding to the maximum

power in the segment PSD.

 P

: Median frequency of the segment PSD.

 P

: Variance of the segment PSD.

 P

: Maximum absolute difference between the

segment PSD and the reference PSD.

These features collectively reflect both absolute

and relative spectral characteristics, providing a

robust representation of the signal’s frequency

content and its deviation from baseline patterns

(Ikizler and Ekim, 2025a, Ikizler and Ekim 2025b).

2.4 Classification and Performance

Evaluation

A Random Forest (RF) classifier was employed to

distinguish between different EEG classes based on

the extracted features. RF is an ensemble learning

method known for its robustness in overfitting and its

ability to handle high-dimensional, complex data

structures. Its use in EEG signal classification has

been well documented due to these advantages (Kode

et al., 2024, Kunekar et al., 2024).

The classification performance was assessed

using standard metrics, including accuracy and

precision, across both binary and multi-class

classification tasks. These metrics provide a reliable

indication of the model's ability to correctly identify

seizure-related activity and differentiate it from non-

seizure EEG patterns (Farawn et al., 2025).

All feature extraction, segmentation, and

classification procedures were implemented in

MATLAB within a fully automated framework,

Impact of Signal Segmentation on EEG-Based Seizure Detection: A Comparative Time-Frequency Analysis

307

ensuring consistency and reproducibility across all

experiments. To ensure a reliable and unbiased

evaluation of the proposed classification framework,

a 10-fold cross-validation strategy was adopted in all

experiments. In this approach, the dataset was

randomly partitioned into 10 equal-sized folds, with

each fold serving as a test set exactly once while the

remaining nine folds were used for training. This

process was repeated iteratively to guarantee that all

data samples contributed to both training and testing,

providing a comprehensive estimate of the model's

generalization ability. The reported accuracy and

precision results represent the average performance

across all 10 folds.

3 RESULTS

In this study, the publicly available EEG dataset

provided by the University of Bonn was utilized to

evaluate the impact of data segmentation on the

classification performance of epileptic seizure

detection. The dataset consists of five distinct classes

(A, B, C, D and E), containing both healthy and

epileptic EEG recordings. Various binary and multi-

class classification tasks were designed by combining

different subsets of these classes to comprehensively

assess the system's effectiveness.

All experimental procedures, including signal

preprocessing, segmentation, feature extraction,

classification, and performance evaluation, were

implemented entirely in a MATLAB environment

using a custom-developed program. This program

was designed to perform the entire experimental

workflow in a fully automated manner, ensuring

consistency and repeatability across all tests.

In the experimental setup, the effect of signal

segmentation was investigated by dividing each EEG

recording into 1, 2, 4, 8, and 16 equal-length

segments. Two widely used feature extraction

techniques were employed separately for each

scenario: DWT and PSD. The extracted features were

subsequently classified using a Random Forest

algorithm, which has been shown to be effective for

EEG based classification tasks due to their robustness

and ensemble learning capabilities.

For each segmentation level and classification

task, both accuracy and precision metrics were

calculated to evaluate the system's performance. The

entire set of experiments, covering 16 different

classification tasks and five segmentation levels for

both DWT and PSD-based feature sets, was

conducted on a personal computer equipped with an

Intel 12th Generation i5 processor and 16 GB of

RAM, running a standard Windows 11 operating

system. The computational environment provided

sufficient processing power to efficiently handle the

relatively large number of experiments without

introducing hardware-related performance

limitations.

The following sections present detailed

experimental results, including quantitative tables

and visual analyses, to reveal the effect of

segmentation on classification accuracy for both

feature extraction approaches.

The effect of data segmentation on the

classification performance was systematically

evaluated using both DWT-based, and PSD-based

feature extraction approaches combined with

Random Forest classification. The detailed results for

each classification task and segmentation level are

presented in Table 1 (DWT) and Table 2 (PSD),

respectively.

In general, increasing the number of segments

applied to the EEG recordings leads to noticeable

changes in classification accuracy. This effect is

evident across both feature extraction strategies,

though with slight differences in magnitude and

behaviour depending on the method.

For the DWT-based feature extraction, the

segmentation process initially contributes positively

to classification performance. Specifically,

segmenting the signals into 2 and 4 parts often results

in improved accuracy compared to the non-

segmented scenario, particularly for complex

classification tasks such as multi-class problems (e.g.,

A-B-C, A-B-C-D-E). However, excessive

segmentation (i.e., 16 segments) tends to degrade

performance, especially in binary tasks such as A-B

and A-C, where a decline in accuracy is observed.

This suggests that excessive division of signals may

disrupt the temporal structure and statistical

characteristics captured by the DWT, negatively

impacting the representational power of the extracted

features.

On the other hand, the PSD-based feature

extraction exhibits a more consistent and stable

improvement trend with increasing segment count. In

particular, the classification accuracy for difficult

tasks such as C-D, C-D-E, and A-B-C-D-E shows

substantial gains as the number of segments

increases. Notably, even at 16 segments, no severe

performance degradation is observed, indicating that

PSD features can benefit from finer temporal

resolution without sacrificing signal integrity. This

can be attributed to the frequency-domain nature of

PSD, which allows for effective characterization of

spectral content even in short signal segments.

ICEEECS 2025 - International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences

308

When comparing the two methods, it is evident

that while both DWT and PSD benefit from moderate

segmentation (2 to 4 segments), PSD-based features

demonstrate greater robustness to higher

segmentation levels, particularly in multi-class and

challenging binary classification scenarios. In

contrast, the DWT-based approach appears to be

more sensitive to over-segmentation, emphasizing the

need to carefully select the segmentation parameter

based on the chosen feature extraction method.

Overall, these results highlight the critical role of

segmentation in optimizing the classification

performance of EEG signals, as well as the interaction

between segmentation strategy and feature

representation. The findings suggest that an optimal

segmentation level exists that maximizes

classification accuracy, and that this optimum may

vary depending on the feature extraction technique

applied.

In addition to the tabular results, Figures 2 and 3

visually illustrate the impact of data segmentation on

the mean classification accuracy across all 16 tasks

for both DWT-based and PSD-based feature

extraction methods, respectively.

Table 1: Classification performance (Accuracy and Precision) of Random Forest for different segment counts and

classification tasks using DWT-based features.

Tasks Metric

DWT-

ase

Feature Vecto

1 segment 2 segment 4 segment 8 segment 16 segment

A-E

Accurac

% 99,50 99,50 100,0 100,0 99,78

Precision % 99,55 99,52 100,0 100,0 99,78

A-D

Accurac

% 96,50 97,25 97,75 96,69 94,34

Precision % 96,89 97,38 97,79 96,74 94,36

A-C

Accuracy % 98,50 97,75 98,12 97,12 94,75

Precision % 98,64 97,85 98,16 97,17 94,79

A-B

Accurac

% 92,00 93,75 92,63 91,75 89,41

Precision % 92,72 94,01 92,80 91,80 89,49

C-D

Accuracy % 68,00 76,50 77,75 77,25 74,97

Precision % 68,95 77,20 78,11 77,34 75,03

C-E

Accuracy % 97,50 98,25 98,37 98,81 98,53

Precision % 97,88 98,38 98,42 98,82 98,54

B-E

Accurac

% 97,50 98,75 99,12 98,25 97,84

Precision % 97,94 98,79 99,15 98,26 97,86

B-C

Accuracy % 97,50 97,50 98,38 98,31 96,97

Precision % 97,73 97,61 98,40 98,35 96,99

B-D

Accurac

% 97,00 98,75 98,00 98,06 97,03

Precision % 97,35 98,83 98,04 98,11 97,05

A-B-C

Accurac

% 92,67 93,83 93,17 92,04 88,67

Precision % 93,23 94,11 93,30 92,18 88,74

A-B-D

Accuracy % 93,00 92,67 93,42 91,58 88,71

Precision % 93,69 93,19 93,69 91,76 88,86

A-B-E

Accurac

% 94,33 94,83 94,08 93,25 91,40

Precision % 94,78 95,07 94,29 93,31 91,48

C-D-E

Accurac

% 73,33 81,83 83,92 83,29 81,33

Precision % 73,29 82,13 84,08 83,29 81,21

A-B-C-E

Accuracy % 94,00 94,25 94,38 93,03 89,97

Precision % 94,71 94,51 94,56 93,14 90,05

A-B-C-D

Accurac

% 78,50 82,38 83,69 82,16 77,73

Precision % 79,08 82,72 83,97 82,20 77,74

A-B-C-D-E

Accuracy % 82,40 84,60 85,60 84,22 79,96

Precision % 82,80 84,63 85,76 84,16 79,86

Impact of Signal Segmentation on EEG-Based Seizure Detection: A Comparative Time-Frequency Analysis

309

Table 2: Classification performance (Accuracy and Precision) of Random Forest for different segment counts and

classification tasks using PSD-based features.

Tasks Metric

PSD-

ase

Feature Vecto

1 segment 2 segment 4 segment 8 segment 16 segment

A-E

Accuracy % 100,0 99,75 99,75 99,94 99,88

Precision % 100,0 99,76 99,76 99,94 99,88

A-D

Accurac

% 99,00 99,25 98,62 97,88 96,37

Precision % 99,09 99,29 98,65 97,89 96,42

A-C

Accuracy % 96,50 98,50 96,62 96,62 93,88

Precision % 96,88 98,55 96,73 96,65 93,92

A-B

Accuracy % 91,50 92,75 91,88 95,63 96,56

Precision % 91,97 93,17 92,05 95,64 96,60

C-D

Accurac

% 83,00 92,75 87,75 90,00 90,75

Precision % 83,91 93,06 87,90 90,10 90,81

C-E

Accuracy % 98,50 99,50 99,63 99,94 99,88

Precision % 98,55 99,50 99,63 99,94 99,88

B-E

Accurac

% 98,50 99,00 99,88 99,88 99,62

Precision % 98,64 99,09 99,88 99,88 99,63

B-C

Accurac

% 97,50 98,75 98,62 99,31 97,94

Precision % 97,63 98,83 98,66 99,32 97,95

B-D

Accuracy % 99,00 99,75 99,12 98,64 98,06

Precision % 99,09 99,76 99,16 98,96 98,08

A-B-C

Accurac

% 91,33 95,33 93,25 95,21 93,50

Precision % 92,65 95,64 93,42 95,29 93,53

A-B-D

Accuracy % 94,67 94,50 93,75 95,52 94,62

Precision % 95,01 94,71 93,93 95,44 94,68

A-B-E

Accuracy % 94,33 95,50 94,92 97,25 97,52

Precision % 94,78 95,74 95,03 97,27 97,54

C-D-E

Accurac

% 87,33 93,00 91,17 93,08 93,23

Precision % 88,34 93,16 91,49 93,09 93,25

A-B-C-E

Accuracy % 93,25 95,88 94,37 96,12 95,00

Precision % 93,63 96,50 94,46 96,19 95,02

A-B-C-D

Accuracy % 86,00 92,50 88,75 90,66 89,73

Precision % 86,39 92,76 88,96 90,72 89,78

A-B-C-D-E

Accurac

% 87,20 93,00 90,15 92,33 91,36

Precision % 88,11 93,37 90,30 92,39 91,39

Figure 2: The effect of the number of segments on the mean

classification accuracy across 16 classification tasks using

Random Forest and DWT-based features.

Figure 3: The effect of the number of segments on the mean

classification accuracy across 16 classification tasks using

Random Forest and PSD-based features.

12 4 8 16

Number of Segments

100

12 4 8 16

Number of Segments

100

ICEEECS 2025 - International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences

310

As observed in Figure 2, the DWT-based

approach exhibits a characteristic trend where the

mean accuracy initially improves with segmentation

but shows a gradual decline beyond a certain point.

Specifically, segmenting the EEG signals into two

parts yields a noticeable improvement in overall

accuracy, suggesting that limited segmentation helps

capture localized temporal patterns more effectively.

However, as the number of segments increases

beyond four, the mean accuracy begins to deteriorate.

This indicates that excessive segmentation may

fragment the temporal structure of the signal,

reducing the ability of the DWT to extract meaningful

features, especially for complex classification tasks.

In contrast, Figure 3 demonstrates a more stable

behaviour for the PSD-based approach. Although

slight fluctuations are present, the mean accuracy

remains consistently high across different

segmentation levels, with the best performance

generally achieved between two and eight segments.

This suggests that the PSD method, being inherently

focused on frequency domain information, is less

sensitive to signal segmentation and can maintain

high classification performance even with finer

temporal resolution. Furthermore, the relatively flat

accuracy curve indicates that PSD-based features are

more robust to variations in the segmentation

parameter compared to DWT-based features.

These graphical results confirm that while

segmentation is a valuable strategy for enhancing

classification performance, its optimal configuration

depends significantly on the chosen feature extraction

method. The DWT method benefits from moderate

segmentation but is more vulnerable to over-

segmentation, whereas the PSD approach

demonstrates greater resilience across a wider range

of segmentation levels.

4 DISCUSSION

The results of this study provide important evidence

regarding how the segmentation strategy directly

shapes the performance of seizure detection systems

utilizing EEG signals. While the technical aspects of

the experimental design are described earlier, it is

crucial to emphasize the broader implications of the

observed trends.

The most striking finding is the clear dependence

of classification success on the interplay between

segmentation and feature extraction technique. The

DWT-based method exhibited notable sensitivity to

the segmentation parameter. Moderate segmentation

levels contributed positively by enhancing the

system's ability to capture transient patterns

characteristic of epileptic seizures. However,

excessive segmentation led to performance

degradation, highlighting a potential trade-off

between temporal resolution and the preservation of

signal integrity.

On the other hand, the PSD-based approach

demonstrated greater stability across different

segmentation levels. The ability to extract consistent

spectral information even from short signal segments

explains the more gradual variations in classification

accuracy observed in this method. This robustness

makes PSD-based features particularly attractive for

real-time seizure detection applications, where short

analysis windows and rapid decision-making are

required.

These findings carry direct implications for

practical, clinically oriented EEG monitoring

systems. Particularly in portable or continuous

monitoring setups, signal segmentation becomes

inevitable due to hardware limitations, memory

constraints, or the need for prompt seizure detection.

The results suggest that careful selection of

segmentation parameters, aligned with the

characteristics of the chosen feature extraction

approach, can maximize detection reliability without

sacrificing system efficiency.

Furthermore, the observed differences between

DWT and PSD approaches highlight that there is no

universal segmentation strategy suitable for all signal

processing pipelines. Instead, a task-specific

optimization process is required, especially for

systems intended for deployment in critical care

environments where false positives or delayed

detections may have severe consequences.

5 CONCLUSIONS

This study provides a comprehensive analysis of how

EEG signal segmentation influences seizure detection

performance, offering valuable insights for the

development of reliable, real-world clinical decision

support systems.

The findings demonstrate that segmentation is not

merely a technical preprocessing step but a decisive

factor that interacts with the feature extraction

strategy to shape classification success. Moderate

segmentation enhances performance, particularly for

methods that exploit time-frequency characteristics,

such as DWT. Meanwhile, PSD-based approaches

offer greater flexibility and resilience to

segmentation, making them promising candidates for

continuous, real-time monitoring scenarios.

Impact of Signal Segmentation on EEG-Based Seizure Detection: A Comparative Time-Frequency Analysis

311

These insights lay the groundwork for future

research directions. Moving forward, extending the

analysis to more heterogeneous and clinically

realistic EEG datasets is essential to validate these

findings under practical conditions. Furthermore,

incorporating advanced deep learning architectures

capable of learning optimal segmentation schemes

adaptively, rather than relying on fixed segment

counts, may yield further improvements in both

accuracy and system efficiency.

In addition, future work should investigate the

trade-offs between segmentation, classification

performance, and computational cost to ensure that

proposed methods are not only effective but also

suitable for deployment in low-power, wearable, or

mobile seizure detection platforms. Ultimately, this

line of research contributes to the development of

more accessible, accurate, and patient-friendly

solutions for epilepsy monitoring and management.

REFERENCES

Almahdi, A. J., Yaseen, A. J., & Dakhil, A. F. (2021). EEG

signals analysis for epileptic seizure detection using

DWT method with SVM and KNN classifiers. Iraqi

Journal of Science, 54-62.

Andrzejak, R. G., Lehnertz, K., Mormann, F., Rieke, C.,

David, P., & Elger, C. E. (2001). Indications of

nonlinear deterministic and finite-dimensional

structures in time series of brain electrical activity:

Dependence on recording region and brain state.

Physical Review E, 64(6), 061907.

Al Farawn, A., Al-Khammasi, S., Alhilali, A. H., & Ali, N.

S. (2025). EEG Feature Selection Techniques for

Epileptic Seizure Detection: Performance and

Evaluation Study. International Journal of

Mathematics, Statistics, and Computer Science, 3, 345-

358.

Ikizler, N., & Ekim, G. (2025a). High-Resolution Power

Spectral Density Approaches for Epileptic Seizure

Detection. Politeknik Dergisi, vol. 1, no. 1.

Ikizler, N., & Ekim, G. (2025b). Investigating the effects of

Gaussian noise on epileptic seizure detection: The role

of spectral flatness, bandwidth, and entropy.

Engineering Science and Technology, an International

Journal, 64, 102005.

Kınacı, E. B., H. Bal, and H. Kınacı., (2024). A Novel

Support Regression-Based Classification Approach for

Epileptic Seizure Detection. Politeknik Dergisi, vol. 27,

no. 2, pp. 587-601.

Kode, H., Elleithy, K., & Almazedah, L. (2024). Epileptic

Seizure detection in EEG signals using Machine

Learning and Deep Learning Techniques. IEEE Access.

Kunekar, P., Kumawat, C., Lande, V., Lokhande, S.,

Mandhana, R., & Kshirsagar, M. (2024). Comparison

of different machine learning algorithms to classify

epilepsy seizure from EEG signals. Engineering

Proceedings, 59(1), 166.

Liu, S., et al. (2023). Epileptic Seizure Detection and

Prediction in EEGs Using Power Spectra Density

Parameterization.” IEEE Transactions on Neural

Systems and Rehabilitation Engineering, vol. 31, pp.

3884-3894.

Milligan, T. A. (2021). Epilepsy: A Clinical Overview. The

American Journal of Medicine, vol. 134, no. 7, pp. 840-

847.

Naidu, G., T. Zuva, & E. M. Sibanda. (2023). A Review of

Evaluation Metrics in Machine Learning Algorithms. In

Computer Science On-Line Conference, Springer

International Publishing, pp. 15-25.

Siddiqui, M. K., et al. (2021). A Review of Epileptic

Seizure Detection Using Machine Learning Classifiers.

Brain Informatics, vol. 7, no. 1, p. 5.

Subekti, S., Indriyanto, S., Rizal, A., Triwiyanto, T., &

Ziani, S. (2024, September). Epileptic Seizure

Detection using DWT-based on MRMR Feature

Selection Method. In 2024 International Conference on

Electrical and Information Technology (IEIT) (pp. 84-

89). IEEE.

Thangavel, P., Thomas, J., Sinha, N., Peh, W. Y., Yuvaraj,

R., Cash, S. S., & Dauwels, J. (2022). Improving

automated diagnosis of epilepsy from EEGs beyond

IEDs. Journal of neural engineering, 19(6), 066017.

World Health Organization (WHO). Accessed 26.05.2025,

https://www.who.int/news-room/fact-

sheets/detail/epilepsy

Zhou, X., Liu, C., Yang, R., Zhang, L., Zhai, L., Jia, Z., &

Liu, Y. (2024). Learning robust global-local

representation from EEG for neural epilepsy detection.

IEEE Transactions on Artificial Intelligence.

ICEEECS 2025 - International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences

312