Predicting Patient-Specific Epileptic Seizures From Scalp EEG Signals Using KNN Model With Transfer Learning

Epilepsy is a prevalent neurological condition that poses a challenge to both patients and physicians. Prompt and accurate prediction of seizures improves patients’ quality of life and outcomes. This study proposes a transfer-learning-based k-nearest neighbor (KNN) approach that employs sophisticated pre-processing techniques and extensive electroencephalogram (EEG) signal data to predict epileptic episodes. Robust and dependable algorithms are required for predicting epileptic seizures. Traditional methods are beneficial because they are restricted in terms of their accuracy and applicability. The transfer-learning-based KNN approach for epilepsy prediction, which is renowned for its effectiveness and ease of use in other disciplines, was improved in this study. We constructed a substantial EEG dataset by using our techniques to manage various data formats, absent values, and critical features. Data coherence was ensured using standardized methods. The enhanced KNN model was tested using Random Forest, Decision Tree, Support Vector Machine, and Logistic Regression. Each model was assessed using F1 score, recall, accuracy, and precision. The proposed model demonstrated a satisfactory performance, as evidenced by its precision, recall, and accuracy.