A new study published in EPB: Urban Analytics and City Science demonstrates the power of interpretable machine learning in understanding urban homelessness patterns. Led by Doctoral Student Shengao Yi (MUSA'23) and Assistant Professor of Urban Spatial Analytics Xiaojiang Li the study analyzed over 42,000 homelessness reports across New York City's 1,712 census tracts. Using advanced AI techniques including Light Gradient Boosting Machine models and SHAP analysis, the research revealed that the importance of predictive factors varies dramatically by location type—information and communication services are most predictive in commercial areas, while crime and median income dominate in residential zones. The study identified critical threshold effects, such as sharp increases in homelessness when median rent exceeds $1,800 or income inequality (Gini index) surpasses 0.53.
Expand Image
