Adaptive Stepwise Feature Selection Approach for EEG-Based Epileptic Seizure Classification.

Area of Science:

• Neurology
• Biomedical Engineering
• Computer Science

Background:

• Advancements in feature selection (FS) optimization algorithms impact epileptic seizure classification.
• Integrating these algorithms into machine learning (ML) models can lead to time complexity, hindering clinical deployment.

Purpose of the Study:

• To propose an innovative adaptive stepwise FS method for epileptic seizure detection (ESD).
• To optimize ML models for improved performance and reduced computational load in seizure classification.

Main Methods:

• Applied discrete wavelet transform (DWT) to preprocess signals and extract linear/nonlinear features.
• Utilized minimum relevance, maximum redundancy (mRMR) for initial feature ranking.
• Implemented a stepwise FS approach to optimize Random Forest (RF), K-Nearest Neighbour (KNN), and Support Vector Machine (SVM) classifiers.
• Validated the method on the CHB-MIT dataset using a patient-independent approach.

Main Results:

• The proposed stepwise FS method integrated with mRMR improved ML model performance.
• Random Forest (RF) achieved the highest performance with 87.69% accuracy, 91.53% sensitivity, and 83.86% specificity using 12 features.
• The proposed stepwise feature selection method (PSFS) demonstrated comparable performance to generalize forward feature selection (GFFS) with significantly less computation.

Conclusions:

• The proposed adaptive stepwise FS method is efficient and effective for epileptic seizure classification.
• This method offers a practical solution for clinical deployment by addressing time complexity issues.
• The study highlights the potential of optimized feature selection in enhancing ML-based neurological disorder detection.