Context

Crime rates are increasing more and more, especially residential burglaries. This thesis includes a study of the Kernel Density Estimation algorithm, and how to use this algorithm for mapping crime patterns based on geographical data. By visualizing patterns as spatial hotspots, law-enforcements can get a better understanding of how criminals think and act. 

Objectives

The thesis focuses on two experiments, including measuring the accuracy and performance of the KDE algorithm, as well as the analysis of the amount of crime data needed to compute accurate and reliable results. 

Methods

A Prediction Accuracy Index is used to effectively measure the accuracy of the algorithm. The development of a Python test program, which is used for extracting and evaluating the results is also included in the study.  

Results

The data from three geographical areas in Sweden, including Stockholm, Gothenburg and Malmoe are analyzed and evaluated over a time period of one year. 

Conclusions

The study conclude that the usage of the KDE algorithm to map residential burglaries performs well overall when having access to enough crimes. The minimum number of crimes for creating a trustworthy hotspot are presented in the result and conclusion chapters. The results further shows that KDE performs well in terms of execution time and scalability. Finally the study concludes that the amount of data that was available for the study was not enough for producing highly reliable hotspots.