An interpretable deep learning framework for early detection of depression using hybrid architectures

Abstract

Current techniques for detecting depression are labor-intensive and subjective, depending on clinical interviews or self-reports. There is a growing adoption of machine learning (ML) and natural language processing (NLP) to automatically identify depression in textual data. The lack of interpretability, which is essential for healthcare applications, is still a major obstacle, though. By combining convolution neural network (CNN) for feature extraction, bidirectional long short-term memory (BiLSTM) for capturing sequential dependencies, and transformer-based pre-trained language model (PTLM) for contextual understanding, this study offers an interpretable framework for early depression identification. Additionally, the system uses a novel interpretability method to guarantee transparent decision-making. The outcome of the proposed system is found to achieve 96.2% accuracy, 94.5% precision, 95.1% recall, and 94.8% F1-score, which is a significant improvement over current method. This framework acts as a useful tool for early mental health intervention.