NeuroWell AI: A Hybrid Deep Learning Framework for Early Detection of Mental Health Risks

Abstract: Mental health disorders represent a growing global crisis, with the World Health Organization estimating that over one billion individuals worldwide are affected by neurological and psychiatric conditions. Early and accurate detection of mental health risks remains a significant challenge due to the multifaceted, heterogeneous, and often latent nature of symptom manifestation. This paper proposes NeuroWell AI, a novel hybrid deep learning framework that integrates Bidirectional Long Short-Term Memory (BiLSTM) networks, Convolutional Neural Networks (CNN), and a Transformer-based attention mechanism to detect early-stage mental health risks from multimodal data sources, including clinical text records, social media posts, physiological signals, and standardized psychiatric questionnaire responses. The framework employs a fusion strategy combining feature-level and decision-level integration across modalities to improve discriminative power. An Explainable AI (XAI) module using SHAP (SHapley Additive exPlanations) is incorporated to enhance clinical interpretability. Experimental evaluation on four benchmark public datasets — CLPsych, DAIC-WOZ, MODMA, and Reddit Mental Health — demonstrates that NeuroWell AI achieves an average accuracy of 94.7%, precision of 93.8%, recall of 95.1%, and F1-score of 94.4%, significantly outperforming state-of-the-art methods. The proposed system offers a clinically relevant, interpretable, and generalizable solution for population-scale mental health screening.

Keywords: Mental health detection; Hybrid deep learning; BiLSTM; Transformer; Multimodal fusion; Explainable AI; Natural language processing; Affective computing