2020/06/14 by Chaitanya Joshi, Clayton D’Ath, Joshi, Chaitanya +5
Engineering · Medicine · Social Sciences · #62P25 #Applications (stat.AP) #Crime Patterns and Interventions #Data-Driven Disease Surveillance #FOS: Computer and information sciences #G.3 #Traffic and Road Safety
paper · pdf · doi:10.48550/arxiv.2006.08008
openalex publication_date 2020/06/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Developing spatio-temporal crime prediction models, and to a lesser extent,\ndeveloping measures of accuracy and operational efficiency for them, has been\nan active area of research for almost two decades. Despite calls for rigorous\nand independent evaluations of model performance, such studies have been few\nand far between. In this paper, we argue that studies should focus not on\nfinding the one predictive model or the one measure that is the most\nappropriate at all times, but instead on careful consideration of several\nfactors that affect the choice of the model and the choice of the measure, to\nfind the best measure and the best model for the problem at hand. We argue that\nbecause each problem is unique, it is important to develop measures that\nempower the practitioner with the ability to input the choices and preferences\nthat are most appropriate for the problem at hand. We develop a new measure\ncalled the penalized predictive accuracy index (PPAI) which imparts such\nflexibility. We also propose the use of the expected utility function to\ncombine multiple measures in a way that is appropriate for a given problem in\norder to assess the models against multiple criteria. We further propose the\nuse of the average logarithmic score (ALS) measure that is appropriate for many\ncrime models and measures accuracy differently than existing measures. These\nmeasures can be used alongside existing measures to provide a more\ncomprehensive means of assessing the accuracy and potential utility of\nspatio-temporal crime prediction models.\n