Keyword: Artificial Inteligence, Spatial Database, Spatial Modeling, Web Developement, Data Analysis
2022
Research
This project developed a spatial risk prediction model for automobile theft in Chicago using 2018 public crime and environmental data. The goal was to identify high-risk areas through geospatial modeling and assess the reliability of hotspot-based methods compared to traditional risk-factor models.
Understanding spatial patterns of crime enables cities to allocate prevention resources more effectively. By identifying potential hotspots before crimes occur, the model supports data-driven urban safety strategies and reduces reliance on reactive enforcement.
A 500×500 ft spatial grid was generated across Chicago and integrated with six environmental risk factors, including proximity to major roads, facilities, and other contextual indicators. Local Moran’s I was used to detect spatial clustering, and Poisson regression models were applied to predict theft counts. Random k-fold and spatial leave-one-group-out cross-validation tested generalizability, while race-context analysis evaluated potential spatial bias.
Developed in 2022 as part of the Master of Urban Spatial Analytics (MUSA) program at the University of Pennsylvania, this project was completed independently under the guidance of Professors Elizabeth Delmelle, Michael Fichman, and Matt Harris.
Models incorporating hotspot features demonstrated greater generalization and lower mean absolute error (MAE = 0.61 vs 0.66) across both random and spatial validations. However, they showed limitations in capturing thefts in the highest-risk zones. The results suggest that hotspot-based spatial models are more transferable across neighborhoods but should be cautiously applied to avoid bias in enforcement contexts.