Machine Learning-Based Feature Extraction Techniques for Epileptic Seizure Detection Using EEG Bio-signals

Abstract

The This paper explores feature extraction algorithms with machine learning strategies to identify epileptic seizures from ElectroEncephaloGram (EEG) signals. Sudden, unpredictable neurological occurrences known as epileptic seizures frequently show up as aberrant electrical activity in the brain. EEG signals offer essential insights into these dynamics, but successful detection requires sophisticated computational methods because of their intrinsic complexity. This study examines feature selection methodologies, which are as follows: correlation-based, information gain, recursive feature elimination, L1 regularization, random forest, principal component analysis, and independent component analysis. This approach enhances the efficacy of diagnostic procedures and facilitates the administration of appropriate therapeutic interventions. These methodologies enable extracting pertinent patterns and features from EEG information. The collected characteristics function as distinctive inputs for machine learning models, facilitating the creation of resilient seizure detection systems. Several machine-learning methods, including Decision Trees (DT), Random Forest (RF), Naive Bayes (NB), K-Nearest Neighbor (KNN), and eXtreme Gradient Boost (XGBoost), are utilized to acquire complex patterns from the retrieved information. The presented methodology demonstrates encouraging outcomes regarding sensitivity, specificity, and accuracy for identifying epileptic episodes. This study enhances non-invasive epileptic seizure detection approaches by integrating feature extraction procedures with machine learning techniques. The findings of this study have considerable implications for expediting intervention and tailoring treatment approaches for persons diagnosed with epilepsy, thereby improving their overall quality of life and well-being.