A Bayesian Belief Network Classifier for Predicting Victimization in National Crime Victimization Survey

This paper presents the development of a Bayes net classifier for prediction of a victimization attribute value for the National Crime Victimization Survey dataset. The National Crime Victimization Survey dataset has over 250 attributes and 216,000 data points, and as such poses a large-scale problem context for classifier development. The classifier was developed using the Weka machine learning software workbench. A set of structural and parameter learning algorithms for the Bayesian belief network were employed in a development effort while ensuring that the computational complexity in both time and space remained within affordable bounds. A number of structural learning algorithms, including local versions of hill-climbing and K2, provided a classification performance of 99% on the testing data. Simulation results indicate that it is feasible to develop a successful Bayesian belief network classifier for the victimization attribute of the National Crime Victimization Survey data.