Automated Major Depressive Disorder (AMDD) Detection using Audio-Visual Feature Discriptor and CNN

Abstract

The severe forms of depression, a widespread mental health issue, can result in self-harm or suicide. Clinical diagnosis and early intervention of depression are greatly aided by an automatic depression detection system. In this study, we suggest a novel automated major depressive disorder (AMDD) detection technique that makes use of audio-visual parameters from patients. PCA is used as dimension reduction for feature selection. After these two types of signals' effective characteristics were extracted and selected, we trained classifiers on each modality. SVM is taken into account for classification. According to our research, fusing features from different data modalities performs better than using just one, and combining audio-visual data modalities' features results in the highest classification accuracy. Comparing the accuracy of the feature selection approach to other methods, PCA significantly increased the accuracy. Also, SVM gives the best in class accuracy of 99.15% on DAIC-WOZ dataset.