Eye State Prediction on Android Devices using Machine Learning for Natural Environment Electroencephalogram...

Area of Science:

• Neuroscience
• Machine Learning
• Mobile Health

Background:

• Electroencephalogram (EEG) signals are crucial for understanding brain activity in medical and cognitive contexts.
• Deep learning models for EEG analysis demand substantial data and computational resources, limiting their real-world application.
• There is a need for efficient EEG analysis methods deployable on resource-constrained devices.

Purpose of the Study:

• To develop a lightweight machine learning pipeline for EEG classification on Android devices.
• To optimize EEG classification for limited data scenarios using TensorFlow Lite.
• To demonstrate the pipeline's effectiveness through a case study of eye state classification.

Main Methods:

• Developed a machine learning pipeline using TensorFlow Lite for Android deployment.
• Collected EEG data from ten participants using the CameraEEG app for eye state classification (eyes open vs. eyes closed).
• Applied artifact removal (Embedded-ASR) and power spectral feature extraction, followed by training a single-channel Support Vector Machine (SVM) model.

Main Results:

• The SVM model achieved 90% accuracy in classifying eye states.
• The model demonstrated robustness across various evaluation metrics.
• The Android application successfully performed EEG classification on smartphones (Google Pixel 7 Pro, Samsung S22).

Conclusions:

• The developed lightweight pipeline enables efficient EEG classification on mobile devices, even with limited data.
• This approach facilitates physiological measurements in naturalistic settings.
• Potential applications include cognitive workload monitoring, seizure detection, and mental health assessment.