Hybrid Crime Prediction Using GRU and ARIMAX Models

Crime prediction and analysis are crucial for enhancing public safety and optimizing law enforcement strategies. This paper proposes a hybrid crime forecasting framework that integrates deep learning and statistical modeling to analyze crime patterns across Indian states and districts. The approach combines a Bidirectional Gated Recurrent Unit (GRU) model and Seasonal Auto-Regressive Integrated Moving Average with Exogenous Variables (SARIMAX) to leverage historical crime data spanning 12 years. The GRU model effectively captures long-term temporal dependencies and complex patterns, while SARIMAX incorporates historical trends and seasonal variations. Their predictions are merged through a weighted hybrid model to enhance forecasting accuracy. To improve predictive performance, feature engineering techniques such as cyclical encoding and MinMax scaling are applied to preprocess temporal and numerical data. The model demonstrates its effectiveness in predicting crime distributions across 30+ categories, including violent crimes, property crimes, and other offenses. Performance is evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> Score, ensuring robust validation. Additionally, visualization tools provide intuitive insights into predicted crime trends, aiding policymakers and law enforcement agencies in resource allocation and preventive measures. This study introduces a novel integration of deep learning and statistical models for crime prediction, offering a data-driven decision-support system to address a critical societal challenge in predictive analytics.