Comparing Machine Learning and Risk Terrain Modeling Approaches in a GIS for Predicting Spatio-temporal Crime Patterns

Authors: Andrew Foy*, Radford University
Topics: Quantitative Methods, Geographic Information Science and Systems, Hazards, Risks, and Disasters
Keywords: Machine Learning, GIS, Modeling, Crime
Session Type: Paper
Day: 4/6/2019
Start / End Time: 8:00 AM / 9:40 AM
Room: Roosevelt 1, Marriott, Exhibition Level
Presentation File: No File Uploaded


There are many techniques that use Geographic Information Systems (GIS) to model and predict crime across space, but many suffer from over prediction and under prediction, or they are not flexible enough to account for time and the non-stationarity of geographic phenomena. Regression, generalized linear modeling (GLM), risk terrain modeling (RTM) and various hot spot/density analytical methods are often used to understand the factors that explain and predict the spatio-temporal patterns of crime. This research compares the results of predictive crime models using MI and RTM methods using retrospective modeling. Multiple years of crime data from Roanoke, Va. and Radford Va. were used to create models that predicted "future" crime, or rather crime that occurred after the training data’s temporal span. There are many advantages to using MI for predictive hot-spot mapping of crime, but there are also limitations. This paper provides a new innovative methodology for understanding and predicting the spatio-temporal patterns of certain crimes and could improve the predictive power of RTM.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login