Machine Learning Based Crime Rate Prediction

Abstract Crime Rate Prediction is an emerging application of artificial intelligence with the potential to strengthen public safety and inform evidence-based policymaking. This represents a machine learning–driven framework for predicting crime trends across Indian states using multi-year, state-level datasets. The system integrates time series modeling (SARIMAX, Prophet) with regularized regression (Ridge Regression) to capture both temporal dynamics and the influence of socio economic factors. Data preprocessing pipelines— including web scraping and cleaning of state-wise crime statistics—were developed to ensure reliable inputs. The framework is deployed as a Flask-based web application, offering interactive visualization and real-time forecasting capabilities. Experimental evaluation demonstrates that hybrid modeling achieves superior accuracy compared to traditional univariate methods, effectively capturing seasonality, trends, and exogenous influences. By providing interpretable predictions and highlighting critical risk factors, serves as a decision-support tool for policymakers and law enforcement agencies. The proposed system can guide resource allocation, enable proactive interventions, and ultimately contribute to reducing crime rates. Future extensions will focus on fine grained district-level prediction, real-time updating, and incorporation of deep learning–based spatio-temporal models.