Classification of Epileptic Seizure Using Machine Learning Algorithms

Epileptic seizure detection remains a critical task in medical diagnosis, with machine learning (ML) algorithms offering promising avenues for accurate classification. This study investigates the efficacy of various ML algorithms in classifying epileptic seizures, focusing on the impact of dataset balance and dimensionality reduction techniques. A balanced dataset of seizure and non-seizure cases was utilized, ensuring robust model training across seizure types and frequencies. Feature extraction was performed using multiple techniques, with a particular emphasis on kernel principal component analysis (KPCA) due to its non-linear transformation capabilities. Classification was subsequently achieved through algorithms including k-nearest neighbors (KNN), random forests (RF), support vector machines (SVM), and decision trees (DT). The result obtained from binary classification scenario with SMOTE, showed the highest accuracies with KNN and RF, each achieving 95.14% with KPCA at d=7. KPCA yielded the most effective results in producing discriminative features for both binary and multi-class classification, highlighting its value for distinguishing seizure from non-seizure cases. These results indicate that a balanced dataset and an appropriate choice of dimensionality reduction—particularly non-linear KPCA—significantly improve classification performance. These findings support the efficacy of combined feature extraction and machine learning approaches in classifying epilepsy-related cases accurately, underscoring their potential in advancing diagnostic tools for epilepsy management.