Selection Of Classifiers For Depression Detection Using Acoustic Features

Depression is an illness that involves the body, mood, and thoughts, and it adversely affects human life. Depression not only lowers the happiness index of individuals but also reduces mindfulness. The increase in the prevalence of clinical depression has been linked to a range of serious outcomes, particularly to an increase in the number of suicide attempts and deaths; making it a public health concern. This underlines the need of an intelligent depression detection system which is able to automatically classify the individual as healthy or depressed. Selection of effective biomarkers plays a vital role in the design of an intelligent depression detection system. For our work, we have used acoustic features extracted from the spontaneous speech samples of the volunteers. By experimenting and evaluating classification results for the dataset of 54 depressed and 75 healthy individuals using different speech features, we found that speech features can be used as a reliable biomarker for depression detection. Speech features like MFCC, pitch, jitter, shimmer and energy have performed better in classifying an individual as a depressed or a healthy one. In the study, the performance of different classifiers like Random Forest, Support Vector Machine (SVM), Gaussian Mixture Model (GMM) and Naive Bayes has been investigated. Among these, hybrid classifier using GMM and SVM has given the best overall classification result.