Multi-Task Classification for Improved Health Outcome Prediction Based on Environmental Indicators.

Area of Science:

• Environmental Health
• Computer Science
• Urban Planning

Background:

• Evaluating neighborhood impacts on health is challenging.
• Google Street View (GSV) offers large-scale environmental assessment potential.
• Manual GSV image annotation is labor-intensive.

Purpose of the Study:

• To develop an efficient method for analyzing neighborhood environments using GSV images.
• To enhance the training of classifiers with limited supervised GSV data.
• To improve the accuracy of predicting built environment indicators and their correlation with health outcomes.

Main Methods:

• Proposed a multi-task classifier leveraging related tasks.
• Utilized readily available Flickr images as an auxiliary dataset.
• Compared multi-task learning performance against single-task learning.

Main Results:

• Multi-task learning improved classifier accuracy, F1 score, and balanced accuracy by up to 6% for environmental indicators.
• The R-squared values for health outcome correlations improved by up to 4% using multi-task learning.
• The proposed method enhances generalization and reduces overfitting with limited data.

Conclusions:

• Multi-task learning is an effective strategy to overcome data limitations in GSV image analysis.
• This approach yields more accurate environmental indicators for health outcome research.
• The method provides a more reliable foundation for understanding neighborhood-environment-health relationships.