Decision Tree Based Depression Classification from Audio Video and Language Information

Abstract

In order to improve the recognition accuracy of the Depression Classification Sub-Challenge (DCC) of the AVEC 2016, in this paper we propose a decision tree for depression classification. The decision tree is constructed according to the distribution of the multimodal prediction of PHQ-8 scores and participants' characteristics (PTSD/Depression Diagnostic, sleep-status, feeling and personality) obtained via the analysis of the transcript files of the participants. The proposed gender specific decision tree provides a way of fusing the upper level language information with the results obtained using low level audio and visual features. Experiments are carried out on the Distress Analysis Interview Corpus - Wizard of Oz (DAIC-WOZ) database, results show that the proposed depression classification schemes obtain very promising results on the development set, with F1 score reaching 0.857 for class depressed and 0.964 for class not depressed. Despite of the over-fitting problem in training the models of predicting the PHQ-8 scores, the classification schemes still obtain satisfying performance on the test set. The F1 score reaches 0.571 for class depressed and 0.877 for class not depressed, with the average 0.724 which is higher than the baseline result 0.700.