Enhancing Depression Detection through 1D Convolutional Neural Networks on DAIC-WOZ Dataset for Investigation of Visual Cues

Abstract

With the increasing prevalence of depression globally, there is a growing demand for advanced, standardized diagnostic tools to assist in early identification and intervention. This work applies deep learning to address the need for standardized and reliable tools for depression identification. In particular, it investigates how one-dimensional Convolutional Neural Networks (1D CNNs) can analyze visual characteristics from the DAIC-WOZ dataset, a collection of clinical interview sessions. The proposed architecture uses a deep CNN designed to discern temporal patterns in facial features. To improve the model's performance, two dropout rates (0.3 and 0.5) and two learning rates (0.001 and 0.0001) are tested. The configuration with a learning rate of 0.0001 and a dropout rate of 0.5 achieves the best overall performance, with a ROC AUC of 0.79 and macro-average precision, recall, and F1-score of 0.79. These results suggest that a higher dropout rate combined with a lower learning rate improves the model's ability to generalize, most likely by reducing overfitting and enabling the model to identify subtle patterns in the data.