Detection & Prediction of Epileptic Seizures Using Machine Learning Model on 3600s Long EEG data

Analysis of epileptic seizure detection and prediction may significantly improve the lives of people living with epilepsy. As the analysis helps to understand the underlying causes of seizures, detection and prediction may help to prevent mishaps and injuries by providing more effective treatment. This paper presents detection and prediction in three stages using the CHB-MIT scalp EEG database. The first stage explores the spectral features of delta, theta, alpha, beta, and gamma bands using wavelet transform and identifies the frequency range of seizure occurrences. The second stage comprises feature extraction, which helps to identify specific patterns and changes in brain activity. One hour EEG recordings of 45 subjects were considered to gain more information about the temporal dynamics of epileptic seizures. In the third stage, machine learning models (SVM, RF, and KNN) were used to classify the data into seizure and non-seizure. Performance was evaluated for training and testing in terms of TPR, NPR, PPV, f-score, and accuracy. Finally, the results and comparison with other recent published work demonstrate the efficacy of the proposed technique for predicting seizures.