Emotional Cities

Can AI predict how urban spaces make you feel?

What makes a place feel good?

We trained deep learning models to predict emotional responses to different site planning layouts. The goal: help designers create cities that enhance human well-being.

SHAP partial dependence plots reveal the relationship between site planning features and emotional responses.

Spatial Heterogeneity

GWR analysis reveals that the same design factor can have opposite effects in different locations. For example, Simpson Index shows strong negative coefficients (blue) in dense urban cores but positive coefficients (red) in peripheral areas, suggesting context-dependent design strategies are essential.

What We Found

Open and visually connected spatial forms—such as line-like slabs and symmetrical layouts—tend to evoke more positive emotional responses. In contrast, enclosed, mono-functional, or overly uniform configurations, like transport buildings or enclosed blocks, may limit emotional stimulation and reduce perceived vibrancy.

Design elements that predict positive emotions:

  • Open spatial forms - Line-like slabs significantly boost positive emotions
  • Symmetrical layouts - Approximate symmetry and axis-guided designs enhance well-being
  • Mixed functionality - Public and residential buildings promote vibrancy over transport facilities
  • Landscape diversity - Higher Simpson Index (moderate levels) correlates with positive emotions
  • Transit accessibility - High accessibility (>2.0) strongly increases emotional responses
  • Low connectivity paradox - Extremely high connectivity may reduce emotional stimulation

Publication: Cao, Q., Wang, W. (corresp.), & Stouffs, R. (2025). Emotional Responses to Site Planning Layouts: A Data-Driven Analysis Using Deep Learning. CAAD Futures 2025, Springer.

Institution: National University of Singapore, 2024-2025

Tech Stack: PyTorch • Computer Vision • Deep Learning • SHAP • GWR