Search

Depression is the most widely affecting of mental illnesses for public health concern. Although there are many treatments for depression, barriers to diagnosis still exist. The intelligent diagnosis relying on extraction of biomarkers provides reliable indicators of depression. This paper proposed a new method of machine learning diagnosis based on vocal source features. The short-term and long-term features were combined for classification and evaluation. The long-term features contained four important short-term features selected by decision trees, and the random forest algorithm and extreme gradient boosting algorithm were used for classification. The results showed that our method was feasible to classify the degree of depression, F1 scores and sensitivity of non-depression were better than traditional short-term features, long-term features, and deep learning approaches. Our study provides a useful tool for preventing and diagnosing early depression.