Explainable Multimodal Deep Learning for Suicide Risk and Psychiatric Disorder Prediction

Abstract

Suicide and psychiatric disorders are significant public health problems globally, and an accurate, timely prediction is needed for early intervention for better clinical outcomes. In this study, an Explainable Multimodal Deep Learning (XMDL) framework is proposed for predicting the suicide risk and psychiatric disorders using multimodal healthcare data from India. The proposed framework combines demographics, clinical, behavioral and textual data by applying multimodal feature fusion, and compares and verifies the performance of various deep learning architectures, such as CNN, LSTM, Bi-LSTM, Transformer and CNN–LSTM hybrid models. The performance of the models was compared by accuracy, precision, recall, specificity, F1-score, Area Under the Receiver Operating Characteristic Curve (AUC-ROC), and confusion matrix analysis. To attain explainability, SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were employed to generate clinically interpretable and transparent predictions.To make predictions transparent and clinically interpretable, Explainability was achieved using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME). The Hybrid CNN–LSTM model outperformed the other models with an accuracy of 96.2% and an AUC-ROC of 0.984. The proposed framework shows that the combination of multimodal deep learning and XAI can provide accurate predictions and good interpretability, which can be used as a valuable clinical decision support system for detecting early risk of suicide and psychiatric disorders in the Indian healthcare system.