DCEM-TCRCN: an innovative approach to depression detection using wearable IoT devices and deep learning

Depression is a psychological disorder that has vital implications for society's health. So, it is important to develop a model that aids in effective and accurate depression diagnosis. This paper proposes a Dynamic Convolutional Encoder Model based on a Temporal Circular Residual Convolutional Network (DCEM-TCRCN), a novel approach for diagnosing depression using wearable Internet-of-Things sensors.

DCEM integrates Mobile Inverted Bottleneck Convolution (MBConv) blocks with Dynamic Convolution (DConv) to maximize feature extraction and allow the system to react to input changes and effectively extract depression-correlated patterns. The TCRCN model improves the performance using circular dilated convolution to address long-range temporal relations and eliminate boundary effects. Temporal attention mechanisms deal with important patterns in the data, while weight normalization, GELU activation, and dropout assure stability, regularization, and convergence.

The proposed system applies physiological information acquired from wearable sensors, including heart rate variability and electrodermal activity. Preprocessing tasks like one-hot encoding and data normalization normalize inputs to enable successful feature extraction. Dual fully connected layers perform classifications using pooled learned representations to make accurate predictions regarding depression states.

Experimental analysis on the Depression Dataset confirmed the improved performance of the DCEM-TCRCN model with an accuracy of 98.88%, precision of 97.76%, recall of 98.21%, and a Cohen-Kappa score of 97.99%. The findings confirm the efficacy, trustworthiness, and stability of the model, making it usable for real-time psychological health monitoring.