Analysis to Improve Classifier Accuracy in Crime Data Prediction

Crime is a global issue that has the potential to harm a country's social and economic well-being. Controlling crime is an unavoidable step that is necessary for a nation's welfare and long-term development. Exposing crimes and the vulnerable conditions which are constantly impacted by their unlawful activity in this digital world is a herculean task. The difficulty of analysing a huge volume of data about crimes as well criminals is a main task of law enforcing officers. Data mining provides us with a number of practical and convenient methods for evaluating large and varied sets of data. It assists organisations and users in uncovering hidden data from a massive database of crime records in order to investigate, control and prevent crime. By applying a blend of computer science and criminal justice, a data mining process can be created which can help in solving crimes more rapidly. The primary goal of this paper is to discuss various classification algorithms, namely decision tree, random forest, and multilayer perceptron and their role in classifying crime data. The work examines crime characteristics of communities and crime data set taken from UCI machine learning repository. To improve accuracy, pre-processing techniques like missing value replacement and manual feature selection based on human expertise are applied to the crime data set. The results of experiments for the classifier models include accuracy, precision, sensitivity, and F-Measure. According to the results, random forest outperforms all other classifiers in predicting criminal activities.