A VIKOR-Driven Deep Learning Framework for Classification of Focal and Non-Focal EEG Signals

Abstract

Accurate classification of focal and non-focal epilepsy is essential for guiding personalized treatment strategies. This study presents a deep learning framework for electroencephalogram (EEG)–based epilepsy classification that integrates systematic preprocessing, multiple neural architectures, and rigorous statistical evaluation. Signal denoising was optimized using the VIKOR (VlseKriterijumska Optimizacija I Kompromisno Resenje) multi-criteria decision-making method, which objectively identified median filtering as the most effective technique for enhancing EEG clarity. Four neural models—Artificial Neural Network (ANN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) were trained and compared using 10-fold cross-validation to ensure reliable and generalizable performance. Among these, CNN consistently achieved the best results, reaching 98% accuracy by effectively capturing morphological patterns and temporal dynamics of the signals. The robustness of these findings was confirmed through one-way ANOVA followed by Tukey's HSD post hoc analysis, which demonstrated that CNN's improvements were statistically significant (p < 0.05). The novelty of this work lies in combining decision-theoretic filter selection, systematic benchmarking of deep architectures, and rigorous statistical validation, contributing to the development of reproducible and clinically relevant frameworks for real-time epilepsy detection.