Evaluation of Feature Selection Methods for Classification of Epileptic Seizure EEG Signals

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Epilepsy significantly impacts patient quality of life and is a common neurological disorder.
• Accurate seizure classification and prediction are crucial for patient management.
• Electroencephalographic (EEG) data and machine learning are widely used for seizure analysis.

Purpose of the Study:

• To compare the performance of various feature selection methods for ictal epilepsy seizure detection.
• To evaluate the impact of different feature sets and subset sizes on classification accuracy.
• To assess the similarity of feature subsets selected by different methods across various classifiers.

Main Methods:

• Utilized the CHB-MIT and Siena Scalp EEG databases for seizure detection.
• Applied and compared multiple feature selection techniques.
• Evaluated classification performance using models such as K-nearest neighbor and random forest.
• Analyzed classification results based on F1-score.

Main Results:

• The K-nearest neighbor classifier with the CHB-MIT dataset achieved the highest F1-score of 0.90.
• No single feature selection method consistently outperformed others across all scenarios.
• Classification performance was significantly influenced by the choice of classifier, dataset, and feature set.
• Optimal classifier/feature selection combinations included K-nearest neighbor/support vector machine and random forest/embedded random forest.

Conclusions:

• Feature selection method performance in epilepsy seizure detection is highly context-dependent.
• The optimal approach requires careful consideration of the specific classifier, EEG dataset, and feature characteristics.
• Further research into tailored feature selection strategies for diverse EEG datasets is warranted.