An Explainable EEG-Based Human Activity Recognition Model Using Machine-Learning Approach and LIME.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Electroencephalography (EEG) non-invasively monitors brain activity during tasks.
• Machine learning (ML) shows promise for human activity recognition (HAR).
• eXplainable artificial intelligence (XAI) can interpret ML models, highlighting key EEG features.

Purpose of the Study:

• To assess the feasibility of an EEG-based ML model for categorizing everyday activities (resting, motor, cognitive).
• To use XAI to clinically interpret which EEG features are most influential in HAR models.

Main Methods:

• Collected EEG data from 75 healthy individuals during resting, walking, working, and reading tasks.
• Applied ML models (Random Forest, Gradient Boosting) for activity recognition.
• Utilized LIME (Local Interpretable Model-Agnostic Explanations) for XAI interpretation of EEG spectral features.

Main Results:

• ML models, especially Random Forest and Gradient Boosting, achieved high accuracy in distinguishing activities.
• ML models' activity recognition aligned with known EEG spectral band patterns.
• XAI confirmed the influence of specific EEG spectral features on HAR model predictions.

Conclusions:

• EEG-based ML models, interpreted with XAI, are feasible for HAR.
• This approach can enhance patient recovery monitoring, motor imagery, and applications in the healthcare metaverse and clinical virtual reality.