Applying Spatiotemporal and Demographic Data to Locate Next Crime Location

Geographic profiling is a tool used by law enforcement to predict the location of a serial criminal’s next crime. A typical geographic profile outputs estimated probabilities with the input of time and location of previous crimes. In this paper, we develop a new geographic profile that is able to incorporate demographical observations while remaining an effective predictor. We assume that (1) there are buffer zones around previous crime scenes because the criminal fears capture, (2) there is distance decay as criminals prefer something about the locations where previous crimes were committed, and (3) criminals target potential victims based on income and (4) target areas based on crime rate, which are claims supported by research of serial criminals. In order to find an effective profile, we have combined two models of criminal behavior which predict the location of future crime. First, we compute probability densities using a time-weighted kernel density algorithm, which includes buffer zone and distance decay functions. We test this model and find it does well with respect to a control algorithm. Second, we utilize Markov chains to model the criminal’s attraction to certain neighborhood characteristics: income and crime frequency. These two profiles are combined into one algorithm by utilizing the time-weighted kernel density algorithm to identify locations at high risk of future crime, then applying the Markov model to further refine our predictions. We then apply our model to the notorious case of the Boston Strangler. The predictions of the combination algorithm are compared against the predictions of a standard control algorithm. Our combination algorithm does not do well in predicting future crimes. We discuss the effects of these weaknesses on the expected effectiveness of the model in predicting future crime locations. Included also is an executive summary for the chief of police.