Predictive Crime Analysis in Bangladesh: Integrating Machine Learning and Time Series

Crime Pattern Analysis is a core component of modern law enforcement that helps law enforcers to ensure effective strategies to enhance public safety. This study presents a comprehensive analysis of crime trends in Bangladesh using publicly available crime data covering 2010-2019. The analysis focuses on major crime categories like murder and robbery. To understand how these crimes changed over the years, different types of regression models like XGB, Decision Tree, Gradient-Boosting, etc. were used to predict the number of murder cases. In addition, time series models such as ARIMA and Prophet were applied to fit the temporal progression of murder and robbery cases. The fitted values from these models were compared to the actual case counts to assess their alignment with real-world data. Evaluation metrics such as R squared, RMSE, MSE, and MAE were used to measure prediction accuracy and model performance. Among the regression models ExtraTreesRegressor demonstrated the best prediction accuracy, achieving the highest R squared score of 0.9766, lowest errors with MAE of 31.12, MSE of 2798.43, and RMSE of 52.90.This combined regression and time series analysis offers valuable insights into the crime pattern in Bangladesh and help law enforcers in decision making and strategic planning.