Detecting epileptic seizure with different feature extracting strategies using robust machine learning classification...

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Epilepsy is a neurological disorder characterized by abnormal brain excitability.
• Electroencephalogram (EEG) monitoring is crucial for detecting epileptic seizures.
• Effective epilepsy detection relies on sophisticated feature extraction and machine learning strategies.

Purpose of the Study:

• To investigate diverse feature extraction techniques for EEG-based epilepsy detection.
• To evaluate the performance of novel machine learning classifiers with optimized parameters.
• To improve the accuracy and reliability of epileptic seizure identification.

Main Methods:

• Extracted time-domain, frequency-domain, nonlinear, wavelet-based entropy, and statistical features from EEG data.
• Employed and optimized Support Vector Machines (SVM), K-Nearest Neighbors (KNN), decision trees, and ensemble classifiers.
• Utilized tenfold cross-validation for robust performance evaluation, assessing metrics like accuracy, TPR, NPR, PPV, and AUC.

Main Results:

• Optimized SVM linear kernel and KNN with City block distance achieved 99.5% accuracy, outperforming default parameters.
• SVM demonstrated high separation with AUC values of 0.9991 and 0.9990 at different kernel scales.
• KNN with inverse squared distance weighting showed improved performance across various neighbor settings.
• Distinguishing postictal heart rate from ictal states yielded 100% performance with several machine learning classifiers.

Conclusions:

• Advanced feature extraction and optimized machine learning classifiers significantly enhance EEG-based epilepsy detection.
• The study highlights the superior performance of optimized SVM and KNN models for accurate seizure identification.
• These findings offer a promising approach for more reliable epilepsy diagnosis and patient monitoring.