Implementation of the Frequency Domain Feature Extraction Technique for Machine Learning-Based Seizure Detection

This study presents a novel approach to seizure detection using electroencephalogram (EEG) signals, combining frequency domain feature extraction with correlation-based feature selection. The research aims to enhance the accuracy and efficiency of automated seizure detection systems, which are crucial for timely intervention and management of epilepsy. Our methodology employs frequency domain analysis to extract relevant features from EEG signals, capturing the spectral characteristics associated with seizure activity. To optimize the feature set, we implement a correlation-based feature selection method, which identifies the most informative features while reducing redundancy and computational complexity. We evaluate the performance of three machine learning models: Random Forest, LightGBM classifier, and Naive Bayes classifier. These models are trained and tested on a comprehensive EEG dataset, with their performance assessed using standard metrics including accuracy, sensitivity, and specificity. The results demonstrate the effectiveness of our approach, with all three models achieving exceptional performance. The Random Forest classifier exhibits the highest overall accuracy at 99.57%, with balanced sensitivity and specificity (both 99.5%). The LightGBM classifier follows closely with 99.35% accuracy, 99.57% sensitivity, and 99.13% specificity. The Naive Bayes classifier shows strong performance as well, with 99.28% accuracy, 99.86% sensitivity, and 98.70% specificity. These findings suggest that our feature extraction and selection methods, combined with appropriate machine learning algorithms, can significantly improve the accuracy of seizure detection.