Using machine learning to examine associations between the built environment and physical function: A feasibility...

Area of Science:

• Urban planning and public health
• Geospatial analysis and machine learning

Background:

• Assessing neighborhood design's impact on health is challenged by data limitations.
• Traditional methods struggle to capture environmental characteristics like greenery and building structure.

Purpose of the Study:

• To explore the feasibility of using Generative Adversarial Networks (GANs) to measure neighborhood design.
• To investigate the relationship between the built environment and physical function using machine learning.

Main Methods:

• Utilized Google Street View and aerial imagery from 200 Brisbane neighborhoods (HABITAT project).
• Trained a GAN model on 56,330 street view images to analyze neighborhood characteristics.
• Compared physical function scores (PF-10) aggregated at the neighborhood level.

Main Results:

• Aerial imagery proved inadequate for model training.
• The GAN model successfully identified differences in neighborhood design between high and low physical function areas.
• Key differences noted in urban greenery (tree height) and dwelling structure (building height).

Conclusions:

• Street view imagery, processed by GANs, is a feasible method for measuring neighborhood design.
• Machine learning can detect built environment features influencing physical function.
• Future deep learning applications require diverse and sufficient imagery datasets.