Graph based ensemble classification for crime report prediction

Abstract

Criminology and crime analysis are growing areas of research dealing with huge amount of past crime reports. Classification analysis identifies the crime patterns among the reports and predicts the crime category of new reports. As different classifiers classify a data set with different accuracy, so ensemble classification is the appropriate choice for many applications. The present work demonstrates a graph based ensemble classification approach to predict crime reports more accurately than the individual classifiers. Initially, the crime reports are graphically modelled to find a set of maximal independent subset of features and subsequently, decision tree based classifiers are designed using these subsets. Next, a graph is constructed with generated decision trees as nodes and the same method is applied to find out a set of maximal independent subsets of decision trees, each of which is considered as an initial ensemble classifier. Finally, these classifiers are ensemble using the traditional Boolean algebra properties to generate the final ensemble classifier. The developed model may be helpful for crime analysis by the law enforcement agencies for controlling the criminal activities in the country. Extensive experiments are conducted for evaluating the performance of the proposed method on several crime data sets. The superior performance demonstrates the effectiveness of the developed ensemble classification model and indicates its wide potential applications in crime domain.