Comparison of classifiers using robust features for depression detection on Bahasa Malaysia speech

Early detection of depression allows rapid intervention and reduce the escalation disorder. Conventional method requires patient to seek diagnosis treatment by visiting a trained clinician. Bio-sensors technology such as automatic using speech can be used assist early for detecting remotely those who are at risk. In this research, we focus on Bahasa Malaysia language signals that recorded via subject’s personal mobile devices. Speech recordings from total 43 depressed subjects 47 healthy were gathered online platform with validation according Malay beck inventory II (Malay BDI-II), health questionnaire (PHQ-9) declaration major depressive disorder (MDD) Classifier models compared time-based spectrum-based microphone independent feature set hyperparameter tuning. Random forest performed best male reading 73% accuracy while support vector machine both spontaneous female 74% accuracy, respectively. Automatic has shown promising learning features but larger database is necessary further improving performance.