Feature selection using angle modulated simulated Kalman filter for peak classification of EEG signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Existing electroencephalogram (EEG) peak classification models use varied features, limiting performance across applications.
• Problem-dependent performance necessitates a unified approach to feature selection in EEG analysis.

Purpose of the Study:

• To combine features from existing EEG peak classification models.
• To develop an optimized feature selection method for improved EEG signal analysis.
• To introduce a novel Angle Modulated Simulated Kalman Filter (AMSKF) for feature selection and classification.

Main Methods:

• A novel Angle Modulated Simulated Kalman Filter (AMSKF) was developed as a feature selector.
• A neural network random weight method was integrated as a classifier within the AMSKF technique.
• 11,781 peak candidate samples from three event-related EEG signals (single eye blink, double eye blink, eye movement) of 30 healthy subjects were utilized.

Main Results:

• The proposed AMSKF feature selector successfully identified the optimal combination of features.
• The AMSKF technique demonstrated performance comparable to existing methods in epileptic EEG events classification.
• Validation was performed on a dataset of 11,781 EEG peak candidates.

Conclusions:

• The AMSKF approach offers an effective solution for feature selection in EEG signal processing.
• This method shows promise for enhancing the accuracy and reliability of EEG-based classification tasks.
• The study validates the efficacy of AMSKF in analyzing event-related EEG signals.