Multimodal E2E framework for depression classification: Preliminary results.

Mental disorders are still common throughout the world and depression has an important impact on those people with mental disorders in the means of daily life and economic burden. Thus, early detection of depression becomes vital for patients, clinicians as well as health policymakers. For detection of depression, there is a need for objective assessment approaches besides subjective ones which are currently used by clinicians. In this study, we provided a multimodal model for non-verbal behaviors including text, audio and video to provide objective approach. Two different depression datasets were used to train and test the developed algorithms (Support vector machines and random forest). Results were provided for each combination of modalities for both algorithms and datasets. The highest F1 score (0.56) was obtained by the combination of audio and video modalities. As a conclusion, the importance and effectiveness of usage of multimodality was emphasized and shown by different performance measures. [ABSTRACT FROM AUTHOR]