Data selection in EEG signals classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Electroencephalogram (EEG) signal analysis is crucial for detecting alcoholism.
• Analyzing multi-channel EEG data presents computational challenges, including complex calculations and long execution times.
• Efficient data reduction is needed for practical online EEG analysis applications.

Purpose of the Study:

• To propose and evaluate novel data selection methods for analyzing EEG signals in alcoholics.
• To reduce the volume of EEG data and computational load while maintaining classification accuracy.
• To compare the effectiveness of different data selection techniques in identifying alcoholics.

Main Methods:

• Proposed three data selection methods: Principal Component Analysis based on Graph Entropy (PCA-GE), Graph Entropy (GE) difference channel selection, and mathematical combinations channel selection.
• Utilized J48 decision tree, K-nearest neighbor, and Kstar classifiers for data classification.
• Evaluated methods based on classification accuracy, data reduction percentage, and computation time.

Main Results:

• The PCA-GE method achieved 94.5% classification accuracy using only 29.69% of the data and 29.5% of the computation time.
• The GE difference channel selection method attained 91.67% accuracy with 29.69% of the original data.
• All proposed methods successfully selected representative EEG data without compromising the accuracy of discriminating alcoholics from non-alcoholics.

Conclusions:

• The developed data selection methods are effective for reducing EEG data size and processing time in alcoholism detection.
• PCA-GE and GE difference methods offer significant efficiency gains for online EEG analysis and classification.
• Minimizing data usage while preserving classification accuracy is vital for practical applications in clinical settings.