American
International
University-
Bangladesh

Students & Supervisors

Abstract

This research explores the effects of major political disruptions on crime dynamics, focusing on Bangladesh. Leveraging machine learning techniques, we analyzed how crime behavior changed before and after a significant political event in mid 2024, which led to widespread unrest and a shift in governance. A variety of regression models were trained on monthly crime data preceding the transition, including Linear Regression, Polynomial Regression, XGBoost, LightGBM, Prophet, and Random Forest. Among them, Random Forest produced the best predictive accuracy, with a mean absolute error (MAE) of 2302 and root mean squared error (RMSE) of 3131 on the pre-transition data. The model was then used to forecast crime behavior during the post-transition period. The results revealed a sharp decline in performance, with tolerance-based accuracy falling from 96.9% before the event to 50% after the event. This significant drop highlights a measurable shift in crime dynamics, likely driven by heightened political uncertainty, weakened law enforcement, and changing public behavior. Supporting visualizations, including heat maps and trend comparisons, confirmed alterations in both crime seasonality and intensity. Our findings underscore the importance of incorporating socio-political awareness into predictive policing models, and demonstrate that traditional forecasting techniques may fail when confronted with sudden systemic disruptions. This study contributes to the growing field of AI-driven criminology by illustrating how machine learning can help quantify the societal impact of political instability on public safety.

Keywords

Publication Details

• Type of Publication:
• Conference Name: 2nd IEEE Conference on Computing Applications and Systems (COMPAS 2025)
• Date of Conference: 23/10/2025 - 23/10/2025
• Venue: Islamic University Kushtia, Bangladesh.