On-device single channel EEG classification on Android smartphones using lightweight machine learning models.

Area of Science:

• Neuroscience
• Machine Learning
• Mobile Health

Background:

• Electroencephalogram (EEG) signals are crucial for cognitive and medical applications, but traditional deep learning models require extensive data and computational power.
• Developing lightweight models for on-device EEG classification is essential for mobile deployment and accessibility.

Purpose of the Study:

• To present a machine learning pipeline for efficient, on-device EEG signal classification on Android devices with limited data.
• To enable offline EEG classification for applications like eye-state detection.

Main Methods:

• Collected EEG data from ten participants performing eyes-open and eyes-closed tasks.
• Utilized Embedded-Artifact Subspace Reconstruction (E-ASR) for artifact removal and power spectral feature extraction.
• Trained a Support Vector Machine (SVM) classifier on a single-channel occipital electrode and deployed it on an Android app.

Main Results:

• Achieved 90% accuracy in classifying eye states (open vs. closed) using a single-channel model.
• Model robustness confirmed through precision, sensitivity, specificity, F1-score, and MCC.
• Successfully demonstrated EEG signal classification on Android devices like Google Pixel 7 Pro and Samsung S22.

Conclusions:

• This pipeline enables EEG classification with limited data on Android devices, enhancing physiological measurement in natural settings.
• The approach is extendable to cognitive workload monitoring, seizure detection, and mental health assessment.
• Demonstrates the feasibility of scalable and accessible smartphone-based EEG monitoring for research and clinical use.