Adaptive Deep Learning Models for Multi-Class Neuropsychiatric Disorder Detection

Abstract

Neuropsychiatric disorders such as depression, schizophrenia, bipolar disorder, and anxiety disorder represent a significant global health burden due to their complex and overlapping symptoms. Accurate and early multi-class classification of these disorders remains challenging because of heterogeneous clinical presentations and noisy biomedical signals. Traditional diagnostic approaches rely heavily on subjective clinical assessments, which often lead to delayed or inaccurate diagnosis. This study proposes an Adaptive Deep Learning framework for Multi-Class Neuropsychiatric Disorder Detection (ADL-MNDD). The proposed model integrates adaptive neural architectures capable of dynamically adjusting feature representations based on input variability. It combines convolutional layers for feature extraction, recurrent structures for temporal dependency modeling, and adaptive attention mechanisms for class-specific feature refinement. The model is evaluated using clinical and EEG-based datasets, and performance is measured using accuracy, precision, recall, F1-score, and ROC-AUC across multiple disorder classes. Experimental results demonstrate that the proposed adaptive deep learning model significantly improves classification performance compared to traditional machine learning and static deep learning models. The framework is suitable for real-time clinical decision support systems and mental health monitoring applications.