Crime Prediction Using Machine Learning: A Comparative Analysis

Abstract

In the past few years, Machine Learning (ML) methods have acquired extensive attention from researchers and even criminology practitioners for their potential in crime prediction and crime forecast. To date, many research studies have already been performed on crime prediction using machine learning techniques. Some common methods that have been used include KNNs, Decision trees, Naïve Bayes, and random forest, among others. In this research study, a comparative analysis of 51 research studies has been performed. The results indicate that the supervised learning approach is the most commonly used by researchers. In addition, random forest was the most commonly used method. In various studies, almost all proposed models and methods produced accurate results and outcomes with the exception of one study. In the future, it is critical to evaluate these ML based algorithms in a real-world situation and identify which situations or factors affect their accuracy. At the same time, it is also necessary to identify and evaluate which techniques can be used to improve the accuracy of each machine learning method and which algorithm or method might prove to be the best choice for police departments to use for predicting crime.