Enhanced performance of EEG-based brain-computer interfaces by joint sample and feature importance assessment

Area of Science:

• Neuroscience and Artificial Intelligence
• Brain-Computer Interface (BCI) Systems
• Signal Processing

Background:

• Electroencephalograph (EEG) is crucial for BCI, but raw data quality varies, and feature relevance across brain regions/frequencies is unclear.
• Directly using multi-channel, multi-band EEG features for recognition is suboptimal due to data non-stationarity and feature dimension importance.
• Existing BCI methods struggle to account for sample quality and feature-specific contributions to mental state recognition.

Purpose of the Study:

• To propose a novel model, Joint Sample and Feature importance Assessment (JSFA), for BCI systems.
• To simultaneously assess the importance of EEG samples and features for improved mental state recognition.
• To address data quality variations and differential feature relevance in EEG-based BCI.

Main Methods:

• Developed the Joint Sample and Feature importance Assessment (JSFA) model.
• Incorporated self-paced learning for sample importance assessment.
• Utilized a feature self-weighting technique for feature importance evaluation.
• Evaluated JSFA on SEED-IV (emotion recognition) and SEED-VIG (driving fatigue detection) datasets.

Main Results:

• JSFA effectively identifies and quantifies the importance of individual EEG samples.
• The model successfully determines the relevance of different feature dimensions (brain regions/frequency bands).
• Significant enhancements in recognition performance were observed for both emotion classification and fatigue detection tasks.

Conclusions:

• JSFA provides a robust method for handling EEG data variability and feature importance in BCI.
• The proposed model enhances the accuracy and reliability of mental state recognition in BCI systems.
• JSFA offers a promising approach for optimizing feature extraction and sample selection in EEG-based BCI applications.