Channel Reduction for an EEG-Based Authentication System While Performing Motor Movements.

Area of Science:

• Neuroscience
• Biometrics
• Machine Learning

Background:

• Commercial adoption of biometric authentication is rising, driving interest in electroencephalography (EEG)-based systems.
• Current EEG authentication methods often require specific user-elicited brain activities, impacting comfort and practicality.
• This research explores using minimal, everyday activities for EEG authentication.

Purpose of the Study:

• To investigate the feasibility of EEG-based authentication using a simple, minimal daily activity (lifting a small object).
• To identify the minimum number of EEG channels required for effective authentication, enhancing user comfort and reducing data processing demands.
• To develop a protocol for ranking EEG channels and assessing authentication performance for custom headset designs.

Main Methods:

• Utilized a public dataset of 32-channel EEG data from 12 participants performing a motor task.
• Filtered EEG data into five frequency bands and extracted 12 distinct features.
• Trained a random forest machine learning model and ranked electrode channels by Gini Impurity.

Main Results:

• Authentication performance was maintained using only 14 EEG channels when data was filtered into the Gamma sub-band.
• This achieved accuracy within 1% of systems using the full 32 channels.
• Identified optimal electrode placement over the frontal and occipital lobes.

Conclusions:

• Effective EEG-based authentication is possible with a minimal number of channels (14) and a simple daily task.
• This approach significantly improves user comfort, reduces data processing requirements for real-time applications, and facilitates custom headset design.
• The developed methodology enables performance ranking for different EEG hardware and tasks, paving the way for practical biometric solutions.