Crime Domain Prediction and Trend Analysis Using Random Forest and XGBoost Models

Abstract

Crime analysis and prediction have become important research areas for improving public safety and supporting law enforcement agencies through intelligent data-driven systems. Traditional crime investigation methods mainly depend on manual analysis and statistical techniques, which are often inefficient for handling large-scale and complex crime datasets. This paper presents a Machine Learning Based Crime Pattern Analysis and Prediction system that utilizes Random Forest and XGBoost algorithms to analyze historical crime records and predict crime domains with high accuracy. The proposed system processes crimerelated attributes such as city, victim age, victim gender, weapon used, and crime category to identify hidden crime patterns and generate predictive insights. The system uses approximately 40,160 crime records collected from 29 Indian cities between 2020 and 2024. Various preprocessing techniques including data cleaning, label encoding, feature engineering, and dataset filtering are applied to improve model performance. The Random Forest model performs crime pattern analysis with an accuracy of 91.25%, while the XGBoost model achieves 94.18% prediction accuracy for crime domain forecasting. An interactive Flask-based web application is developed to provide real-time crime analysis, prediction, and visualization through dynamic yearly and monthly crime trend charts. The proposed system improves prediction accuracy, scalability, analytical efficiency, and decision-making capability, thereby assisting law enforcement agencies in proactive crime prevention and intelligent public safety management.