Fusing Location Data for Depression Prediction.

Area of Science:

• Digital Phenotyping
• Mental Health Informatics
• Mobile Sensing

Background:

• Geographic location data from smartphones shows promise in predicting depression.
• Existing methods using GPS data often suffer from significant missing data, limiting analysis.
• Common practice involves discarding incomplete data, potentially losing valuable insights.

Purpose of the Study:

• To develop and evaluate a data fusion approach for smartphone location data (GPS and WiFi association records).
• To assess the impact of fused, more complete location data on depression prediction.
• To explore the utility of incorporating external WiFi infrastructure data.

Main Methods:

• Collected GPS and WiFi association records from 79 college students' smartphones.
• Developed a data fusion technique to integrate data from multiple location sources.
• Compared depression prediction performance using fused data versus traditional methods.

Main Results:

• The data fusion approach significantly increased the completeness of location data.
• Features extracted from fused data showed stronger correlations with self-reported depression scores.
• Depression prediction F1 scores improved substantially, reaching 0.76 compared to 0.5 without fusion.

Conclusions:

• Fusing GPS and WiFi location data from smartphones enhances data completeness for mental health research.
• Improved data quality via fusion leads to more accurate depression prediction models.
• Incorporating external WiFi infrastructure data did not further improve prediction accuracy beyond phone-based fusion.