Time's Effect on Crime Prediction Precision and Accuracy

Abstract

Crime detection with high prediction and accuracy has become a focused issue in the process of crime investigation and prevention. The higher the accuracy or precision of a crime detection model the more efficient crime investigation and prevention becomes. The present research aims to examine the possible precision and accuracy differences of using two different test datasets (TD) to calculate the predictive accuracy index (PAI), and the recapture rate index (RRI) for kernel density estimation (KDE), risk terrain modeling (RTM) and the combined RTM–KDE model. The present study focuses on theft and robbery cases of between 1 December 2015 and 30 November 2018 in Budapest, Hungary. The novelty of the research lies in its first-time usage in Budapest, in a Central European evolutionary urban structure. The results show that there are differences in prediction performance and in each model using a Test Dataset (TD) more distant in time from the initial dataset resulted in a more accurate prediction. The research proved that datasets with different time distances can have an impact on the predictive accuracy and precision of crime detection.