Scopus Indexed Publications

Paper Details


Title
Crime Report Analysis and Future Crime Forecasting Using Machine Learning: A Bangladesh Perspective

Author
, Fakrul islam ,

Email

Abstract

Crime is a growing social and economic challenge in Bangladesh, placing increasing pressure on law enforcement agencies. Existing crime analysis systems rely primarily on descriptive statistics and retrospective reporting, lacking the capability to forecast future crime trends. This paper presents a machine learning-based framework for historical crime analysis and future crime forecasting using official Bangladesh police data (Crime-Stat-BD-20-25) covering January 2020 to December 2025, encompassing 15 crime categories across multiple policing units. The dataset comprises monthly records from official police units including DMP, CMP, and Dhaka Range. Supervised regression models Ridge Regression, Random Forest, Support Vector Regressor (SVR), and a custom Transformer-based forecaster are trained and evaluated on a held-out test set using MSE, MAE, RMSE, and R2 metrics. Ridge and Bayesian Ridge achieve near-perfect internal reconstruction (R2 = 1.00), attributable to the linear summation structure of Total_Crime rather than generalizable predictive power; the Transformer, performing genuine next-month temporal forecasting from sliding windows, attains R2 = 0.742. A chronological train-test split (January 2020–December 2024 for training, January–December 2025 for testing) is employed throughout to ensure valid out-of-sample evaluation. Division-level 12-month forecasts and risk indices are generated to support proactive resource allocation and preventive policing in Bangladesh. Ethical considerations regarding data bias and responsible deployment are also discussed.


Keywords

Journal or Conference Name
2026 International Conference on Power, Electronics, Communications, Computing, and Intelligent Infrastructure, PECCII 2026

Publication Year
2026

Indexing
scopus