Global models and local patterns for crime prediction from weather data

Abstract

Weather influences people indirectly in many ways. Does this include criminal behavior? Previous research has shown a definite relationship between high temperatures and an increase in violent behavior. This thesis attempts to determine and examine more complex relationships with the use of both local and global machine-learning and data-mining models. The resulting global models perform only marginally better than simple baseline models. However, local patterns built with the Patient Rule Induction Method yield interesting subgroups that are in line with preceding research elsewhere.