Eyelid Movement Command Classification Using Machine Learning.

Area of Science:

• Assistive Technology
• Machine Learning
• Human-Computer Interaction

Background:

• The Eyelid Drive System (EDS) is an emerging assistive technology.
• It enables users to control external devices through eye movements like blinking and winking.
• Existing control methods may have limitations in accessibility or resource requirements.

Purpose of the Study:

• To evaluate machine learning classifiers for the Eyelid Drive System (EDS).
• To assess classifier accuracy, computational demands, and transferability across users.
• To identify efficient algorithms for real-time control of assistive devices.

Main Methods:

• Trained four machine learning classifiers on single-subject blink/wink data.
• Validated classifiers on the training subject and two additional subjects.
• Assessed performance based on accuracy, computational resources, and memory usage.

Main Results:

• Support Vector Machine (SVM) achieved 97.5% accuracy but with high resource demands.
• Logistic regression achieved 96.5% accuracy, using significantly fewer computational and memory resources.
• Classifier performance and transferability varied across subjects.

Conclusions:

• Machine learning, particularly logistic regression, shows promise for EDS control.
• Efficient algorithms are crucial for practical implementation of blink-based assistive technology.
• Further research is needed to optimize transferability and reduce resource requirements for broader user adoption.