A novel multi-modal depression detection approach based on mobile crowd sensing and task-based mechanisms.

Area of Science:

• Computational psychiatry
• Machine learning in healthcare
• Digital phenotyping

Background:

• Depression is a global health issue, exacerbated by the COVID-19 pandemic.
• Existing depression detection methods include task-based and Mobile Crowd Sensing (MCS).
• Integrating these methods offers complementary strengths for enhanced detection.

Purpose of the Study:

• To propose a novel, end-to-end machine learning pipeline for depression detection.
• To combine real-time Mobile Crowd Sensing (MCS) and task-based mechanisms.
• To distinguish between depressed and non-depressed individuals using multimodal data.

Main Methods:

• Developed a machine learning pipeline for multimodal data collection, feature extraction, selection, fusion, and classification.
• Created a real-world dataset of depressed and non-depressed subjects.
• Experimented with various features, feature selection techniques (Pearson's correlation), data fusion, and classifiers (Logistic Regression, SVM).

Main Results:

• Combining features from multiple data modalities outperformed single modalities.
• Fusion of features from all three modalities yielded the highest classification accuracy.
• Support Vector Machines (SVM) achieved the best accuracy of 86%.
• Pearson's correlation-based feature selection improved accuracy compared to other methods.

Conclusions:

• The proposed multimodal approach enhances depression detection accuracy.
• Integrating MCS and task-based methods offers a promising direction for digital mental health.
• The findings demonstrate the advantage of multimodal data fusion over single modalities for depression recognition.