Actigraphy-Based Decision Support System for Objective Diagnosis of Depression Using Deep Learning

Mental health issues are of paramount importance, yet their timely detection lacks direct quantitative tests. However, with the advent of wearable IoMT devices like smartphones, wrist actigraphs, and machine learning (ML) algorithms, real-time data can be harnessed to provide efficient quantitative measures for diagnosing various mental health problems. In this context, this paper introduces a deep convolutional neural architecture designed to analyze IoMT-based actigraphy signals for detecting depressive episodes. The proposed approach not only addresses data loss challenges arising from missingness mechanisms through a custom augmentation algorithm but also ensures class balancing using the SMOTE technique. The suggested model achieves 95.81% accuracy using sophisticated deep learning and attention methods. This discovery allows wearable IoMT devices to identify depressive episodes more accurately and reliably, enabling on-demand, real-time mental healthcare services.