A Comparative Analysis of Machine Learning Techniques for Epileptic Seizure Detection and Classification
Indrani Bhattacherjee
Abstract
The extraction and classification of Electroencephalogram (EEG) signals are crucial for accuracy in detecting an epileptic seizure. Accurate feature extraction and classification are crucial for determining the accuracy and efficiency of the process of epileptic seizure detection. In this research, different machine learning techniques are compared for epileptic seizure detection and classification. The epileptic seizure activities are determined by using the University of Bonn dataset. Using Fine K-Nearest Neighbor (K-NN), Weighted K-NN, Fine Gaussian Support Vector Machine (SVM) as classifiers, the Human Activity Recognition method has been adopted. Activities like “Seizure Patient”, “Seizure free”, “Healthy Eye Open”, “Healthy Eye Closed” can be accurately detected. In comparative analysis of performance of machine learning classifiers used, the following results were obtained: Fine KNN method gave 84% accuracy with Holding out validation of 25% and 100% seizure detection. Weighted KNN achieved 78% accuracy withholding out validation of 25% and 100 % seizure detection. Fine Gaussian SVM achieved 74% accuracy with Holding out validation of 25% and 100% seizure detection.