Akademik Veri Yönetim Sistemi
2nd International Conference on Advances in Electrical, Electronics, Energy, and Computer Sciences (ICEEECS2025), Skopje, Makedonya, 24 - 25 Ekim 2025, ss.1-8, (Tam Metin Bildiri)
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.