Skip to main navigation menu Skip to main content Skip to site footer

Explainable Age Predictions from Electroencephalography Data


Electroencephalography (EEG) serves as a widely acceptable clini- cal tool for monitoring and assessing brain activities. In leveraging artificial intelligence, machine learning techniques have been utilized for neurophysiological age prediction from EEG data. This research aims to enhance the transparency of such models, using SHapley Additive exPlanations (SHAP) to evaluate EEG feature significance. We employ EEGNet for feature extraction, training predictive mod- els like Random Forest, Support Vector Regressor, and a Recurrent Neural Network. Additionally, we incorporate a transformer model for improved clarity and performance evaluation. Our data, sourced from public hospitals, indicates that the Transformer model notably excels in age prediction. This finding underscores the potential for more transparent machine learning in clinical EEG analysis. Our study advances the search for more interpretable and accountable mod- els in healthcare, addressing trust concerns and facilitating informed decision-making in brain health assessment.