Crime Prediction in Chicago: Using a Statistical Learning Approach to Create a Spatiotemporal Vulnerability Surface of Violent Crime
Timmy Huynh, Pennsylvania State University
Brian Swedberg, Pennsylvania State University
Much research has analyzed spatial, demographic, or weather-related explanators for violent crime in cities. In our analysis, we combined all these factors into a single time-dependent surface of the likelihood of violent crime occurrences in the Near West Side neighborhood in the city of Chicago. Since we are able to spatially link together data from multiple sources (e.g., City of Chicago, U.S. Census, and NOAA), each data point in our analysis contains information for all our predictor variables. However, because our independent variable (violent crime occurrence) happens at a fixed point and is spatially-sparser than our other data, we used a distance-decay formula to interpolate the violent crimes over our surface. We then trained and tested a random forest classifier to create a final violent crime vulnerability surface over our selected neighborhood. Our accuracy of violent crime prediction proved to be fairly good for a social science problem.