An Analytical Comparison of Crime Prediction using Machine Learning Techniques

Predicting crime rates is a crucial part of public safety, and improving predictive capacities calls for interdisciplinary work. This work uses multi-year datasets from Chicago to investigate the use of machine learning algorithms in crime rate prediction. Five machine learning models (Polynomial Regression, Random Forest, LightGBM, XGBoost, and Linear Regression) are carefully cleaned and analyzed in order to determine how well they predict crime patterns. The results show that XGBoost Regression performs better than the other models, exhibiting better prediction accuracy and data-generalization capabilities. Conventional linear models such as Polynomial and Linear Regression reveal limits in capturing the intricacy of crime data, however LightGBM and Random Forest Regression also demonstrate encouraging results. The study emphasizes the significance of choosing suitable machine learning algorithms for tasks related to crime prediction and points to potential directions for further investigation, such as fine-tuning models and investigating sophisticated algorithms. In general, the results offer significant perspectives to the discipline of criminology, with the goal of augmenting public safety and facilitating the advancement of more efficacious approaches to deterring crime.