Abstract
Epileptic seizure detection using electroencephalography (EEG) offers a non-invasive means to monitor brain activity and detect neural anomalies. However, extracting detailed time and frequency features from EEG signals is challenging due to their inherent complexity. This study introduces a detection framework leveraging discrete wavelet transform (DWT) to extract robust features across multiple frequency bands, followed by dimensionality reduction to create compact and informative feature representations by using sophisticated feature extraction techniques such as principal component analysis (PCA), independent component analysis (ICA), and DWT. These methods retain essential signal characteristics, which are subsequently classified using K-nearest neighbor (KNN), Naive Bayes (NB), or support vector machine (SVM). Empirical evaluation on the Bonn EEG dataset reveals that the linear discriminant analysis (LDA)-NB configuration achieved a perfect accuracy of 100%, outperforming other combinations, including LDA-SVM at 89.17%, LDA-KNN at 80.42%, PCA-NB at 89.92%, and ICA-SVM at 90.42%. The LDANB model’s ability to achieve optimal performance across all metrics underscores its effectiveness for seizure detection. Most of the existing work has used a large dimension of feature space, which indirectly increases the time complexity. However, in this study, we have integrated feature extraction with feature selection to reduce the dimensionality of the features. The proposed method can detect epilepsy with very high accuracy.