Digital Twin Framework for Mental Health Prediction Using Neural Networks and Explainable AI

Abstract

Digital Twin technology offers a promising avenue for personalized healthcare by creating virtual replicas of patients to enable continuous monitoring and predictive analytics. This paper presents a Digital Twin–based framework for mental health risk prediction using physiological signals (EEG frequency bands and Galvanic Skin Response). A synthetic dataset of 10,000 samples was used to train a Multilayer Perceptron (MLP) classifier; evaluation on a held-out test set (2,000 samples) produced an accuracy of approximately 92.5% and AUC ≈0.98. SHAP explainability was integrated to identify influential features. Results are compared with baseline Logistic Regression and Random Forest models. We discuss implementation details, results, and future work toward clinical deployment of mental health Digital Twins.