Harnessing Creative Methods for EEG Feature Extraction and Modeling in Neurological Disorder Diagnoses

Abstract

Amidst the rising incidence of neurological disorders detected through electroencephalograms (EEG), this study explores innovative techniques for feature extraction. Analyzing EEG data from 88 subjects across 19 channels, a diverse set of 18 features including Relative Intensity Ratio, Power Spectral Intensity, Petrosian Fractal Dimensions, Hjorth Mobility, Hjorth Complexity, Detrended Fluctuation Analysis, Higuchi Fractal Dimension, Hjorth Activity, Sample Entropy, and Lempel-Ziv Complexity and many more are extracted. Encompassing temporal and spectral domains, these features provide comprehensive insights into neurophysiological processes, enabling nuanced EEG data exploration and identification of subtle patterns linked with various neurological disorders. Through rigorous analysis, we evaluate the efficacy of these features in precise disease discrimination using advanced building on two models: Bagging Blended Combination of XGBoost and LightGBM (BBE-XL) and a Multilayer Artificial Neural Network (ML-ANN). By deciphering intricate EEG signal information, this study aids in early detection and intervention for EEG-related disorders, with 97.62% accuracy.

Key Findings

• Achieved 97.62% diagnostic accuracy using a novel Bagging Blended Combination of XGBoost and LightGBM (BBE-XL).
• Extracted 18 diverse features across temporal and spectral domains, including Fractal Dimensions and Hjorth parameters.
• Analyzed EEG data from 88 subjects across 19 channels to identify subtle patterns linked to neurological disorders.
• Demonstrated that combining creative feature extraction with ensemble learning significantly improves disease discrimination.