Feature Selection Using F-statistic Values for EEG Signal Analysis.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) generates complex, non-stationary signals reflecting cortical activity.
• Accurate analysis and feature selection from EEG are crucial for applications like epileptic seizure detection.
• Existing methods may not fully leverage the discriminative power of combined feature sets.

Purpose of the Study:

• To enhance EEG classification performance by developing a novel feature selection approach.
• To identify and select the most discriminative features from a combined set of frequency and entropy-based measures.
• To improve the accuracy of epileptic seizure detection using selected EEG features.

Main Methods:

• Extracted nine features per EEG channel: six sub-band spectral powers and three entropy measures (sample, permutation, spectral).
• Ranked features across all channels using F-statistic values to identify discriminative components.
• Utilized Support Vector Machine (SVM) classification with the selected features.

Main Results:

• The proposed method achieved high performance metrics on the CHB-MIT dataset.
• Average sensitivity reached 92.63%, demonstrating effective detection of seizure events.
• Average specificity was 99.72%, indicating a low rate of false positives.
• The F-1 score averaged 91.21%, reflecting a robust and balanced classification performance.

Conclusions:

• The developed feature selection strategy significantly enhances EEG-based classification.
• Combining frequency and entropy features with F-statistic ranking improves discriminative power for seizure detection.
• The approach demonstrates high efficacy and potential for clinical application in epilepsy monitoring.