The effects of day-to-day variability of physiological data on operator functional state classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Pattern classification techniques are increasingly applied to physiological data, including magnetic resonance imaging and electroencephalography (EEG).
• Applications range from disease diagnosis and brain-computer interfaces to decoding brain activity and enhancing cognitive task performance.
• Ensuring stability of these classifiers across repeated sessions is crucial for generalizable results and practical system development.

Purpose of the Study:

• To evaluate the stability and performance of three popular pattern classification techniques applied to EEG data across multiple days.
• To investigate the impact of cross-day classification on classifier accuracy.
• To introduce and assess modifications aimed at improving cross-day classification stability.

Main Methods:

• Three distinct pattern classification algorithms were applied to EEG data.
• Data were collected from subjects performing a complex multitask over five days within a one-month period.
• Performance was assessed above chance levels, and the impact of cross-day classification was analyzed. Modifications were implemented and tested.

Main Results:

• All three classifiers performed significantly above chance levels.
• Cross-day classification led to a significant decrease in performance for all classifiers.
• Two proposed modifications substantially reduced misclassifications, improving cross-day stability.

Conclusions:

• Pattern classification of EEG data is a viable approach for various applications.
• While cross-day classification presents challenges, proposed modifications enhance stability.
• These findings suggest that pattern classification techniques can be reliably used across days and weeks with appropriate methods.