Mental disorder assessment in IoT-enabled WBAN systems with dimensionality reduction and deep learning

Abstract

Mental health is an important aspect of an individual’s overall well-being. Positive mental health is correlated with enhanced cognitive function, emotional regulation, and motivation, which, in turn, foster increased productivity and personal growth. Accurate and interpretable predictions of mental disorders are crucial for effective intervention. This study develops a hybrid deep learning model, integrating CNN and BiLSTM applied to EEG data, to address this need. To conduct a comprehensive analysis of mental disorders, we propose a two-tiered classification strategy. The first tier classifies the main disorder categories, while the second tier classifies the specific disorders within each main disorder category to provide detailed insights into classifying mental disorder. The methodology incorporates techniques to handle missing data (kNN imputation), class imbalance (SMOTE), and high dimensionality (PCA). To enhance clinical trust and understanding, the model’s predictions are explained using local interpretable model-agnostic explanations (LIME). Baseline methods and the proposed CNN–BiLSTM model were implemented and evaluated at both classification tiers using PSD and FC features. On unseen test data, our proposed model demonstrated a 3–9% improvement in prediction accuracy for main disorders and a 4–6% improvement for specific disorders, compared to existing methods. This approach offers the potential for more reliable and explainable diagnostic tools for mental disorder prediction.