Explaining urban street perception inequities between residents and tourists using interpretable machine learning

Area of Science:

• Urban Planning
• Computer Science
• Environmental Psychology

Background:

• Inclusive urban design requires understanding diverse user perceptions of street environments.
• Existing methods often lack scalability or interpretability in analyzing these perceptions.

Purpose of the Study:

• To develop an interpretable and scalable machine learning framework integrating Street View Images and subjective evaluations.
• To examine perceptual differences in urban street environments between residents and tourists.

Main Methods:

• Collected perception ratings (safety, comfort, convenience, pleasure, sociability) in Xi'an's Mingcheng District.
• Employed a machine learning framework combining predictive modeling and explainable analysis.
• Analyzed the influence of visual and environmental features on perceptions.

Main Results:

• Tourists prioritized symbolic and aesthetic cues, while residents focused on functional and comfort-related features.
• Key visual elements like vegetation, building facades, and spatial openness impacted perceptions differently for each group.
• Identified both linear and nonlinear drivers shaping distinct group perceptions.

Conclusions:

• Revealed significant perceptual disparities between residents and tourists regarding urban street environments.
• Provided actionable insights for developing equitable street design strategies tailored to diverse urban users.
• Highlighted the utility of machine learning for perception-informed urban design.