Comparing the Spatiotemporal Patterns of Crime Incidents based on Time Series Analysis: A Case Study of Austin, Texas

Abstract. As urban populations continue to expand, understanding the patterns of urban crimes becomes increasingly important for public safety and prevention strategies. This study uses spatiotemporal autoregressive integrated moving average (STARIMA) and geographical random forest (GRF) models to explore and compare the dynamics of two distinct categories of crime as classified by the Federal Bureau of Investigation: crimes against persons (e.g., assault, rape) and crimes against property (e.g., burglary, vandalism). While STARIMA is a well-established method for capturing linear spatiotemporal dependencies, GRF integrates spatial heterogeneity and nonlinear relationships through local weighting. Our results show that GRF consistently outperforms STARIMA across both crime types, particularly for property crimes, where broader temporal trends and spatially variable patterns play a stronger role. This work demonstrates the importance of disaggregating crime by type and selecting modeling approaches that accommodate the unique spatial and temporal characteristics of each. The results provide a deeper understanding of crime prediction and guidance for data-driven policing strategies.