A Performance Analysis of Business Intelligence Techniques on Crime Prediction

Abstract

Law Enforcement agencies are faced with a problem of effectively predicting the likelihood of crime happening given the past crime data which would otherwise help them to do so. There is a need to identify the most efficient algorithm that can be used in crime prediction given the past crime data. In this research, Business intelligence techniques considered was based on supervised learning (Classification) techniques given that labeled training data was available. Four different classification algorithms that is; decision tree (J48), Naïve Bayes, Multilayer Perceptron and Support Vector Machine were compared to find the most effective algorithm for crime prediction. The study used classification models generated using Waikato Environment for Knowledge Analysis (WEKA). Manual method of attribute selection was used; this is because it works well when there is large number of attributes. The dataset was acquired from UCI machine learning repository website with a title ‘Crime and Communities’. The data set had 128 attributes of which 13 were selected for the study. The study revealed that the accuracy of J48, Naïve bayes, Multilayer perceptron and Support Vector Machine (SMO) is approximately 100%, 89.7989%, 100% and 92.6724%, respectively for both training and test data. Also the execution time in seconds of J48, Naïve bayes, Multilayer perceptron and SVO is 0.06, 0.14, 9.26 and 0.66 respectively using windows7 32 bit. Hence, Decision Tree (J48) out performed Naïve bayes, Multilayer perceptron and Support Vector Machine (SMO) algorithms, and manifested higher performance both in execution time and in accuracy. The scope of this project was to identify the most effective and accurate Business intelligence technique that can be used during crime data mining to provide accurate results.