pict

BayesNetCrime: System for Analysis of the Factors Affecting Crime

crimenet_small

Crime is an inevitable societal problem and is a major problem of the nation. Effective crime control requires accurate prediction for decision making in crime control planning. This research analyses the factors affecting crime risk, especially, murder cases in the Bangkok Metropolitan Area, Thailand by using a Bayesian Network. The results from the analysis are expected to be used for crime control planning.

The BayesNetCrime System

The Bayesian Network model was developed by expert elicitation and crime theory and it learned using Hugin Researcher 6.3 machine learning software. The factors considered in this study are classified into five main groups: variables describing population, variables describing crime location, variables describing types of crimes, variables describing traffic, and variables describing the environment. Due to the uncertainty and incomplete nature of the variables, Bayesian Network theory is used to analyse the data since it is well suited to dealing with noisy and incomplete crime data.

The data were collected from the National Statistical Office of Thailand, the Royal Thai Police, the Bangkok Metropolitan Administration and the Ministry of Transportation. In this research, data from January 2000 to December 2003 was used.

crimenet

A Bayesian Network model for analysis of the factors affecting crime risk

The result from this analysis can be used to help in crime control planning and environmental design to prevent crime. Based on data from the study, the environmental factor, the number of drug-sale areas in a district, had the most powerful influence on the expected murder rate. By concentrating on the elimination of the drug trade the government could greatly reduce the murder rate. An empirical study on the predictive accuracy performance of the model is included. Receiver Operating Characteristic analysis was used to test the model. The results of the study show the model performed well.

Hugin Software

During the development of the BayesNetCrime System, the Hugin Software package provided a number of benefits as described below.

  • Friendly User Interface. Hugin provides an easily understood user interface environment. This has made the develop work a lot easier and saved a huge amount of time in calculating the probabilities.
  • A Variety of Functionalities. Hugin provides a variety of functionalities that we have found very useful in building the Bayesian network model.
  • The Fastest Decision Engine. The Hugin Decision Engine proved to be the fastest, most efficient, and most reliable inference engine.