Multimodal depression detection using deep learning in the workplace

Stress, followed by depression has become the most common phenomenon in modern work environment. Early detection of depression is important to avoid health degradation and prevent suicide tendencies. Noninvasive monitoring of stress level is effective at screening stage. Many methods based on visual cues, audio feeds and text messages have been used for noninvasive monitoring. In individual modalities, the accuracy is low and false positives are high. This study proposed a multimodal depression classification system based on deep learning. The proposed solution integrates visual, speech and text feeds and extracts deep learning features from these feeds. The features are classified into emotions and temporal emotion variability, and the depression level is classified.