A Novel Fusion Approach for Advancement in Crime Prediction and Forecasting using Hybridization of ARIMA and Recurrent Neural Networks

Introduction: Law enforcement agencies have struggled to forecast crime due to the complexity of criminal conduct and its dynamic elements. Regular developments and the growth of well-known and new platforms have made social media a pervasive way for people to express their ideas and experiences, popularizing blogging and democratizing information distribution. Traditional time series models like Autoregressive Integrated Moving Average (ARIMA) are frequently employed for forecasting, but they may struggle to capture crime data's nonlinear relationships and long-term dependencies. However, Long Short-Term Memory (LSTM) are best at catching sequential patterns but may struggle with short-term data or rapid criminal patterns. Combining strength of both methods may improve forecast accuracy and robustness. Objectives: To develop a hybrid ARIMA-LSTM model that effectively captures both linear temporal trends and complex nonlinear patterns in crime data, enhancing predictive accuracy across diverse crime types and geographic regions in India. Methods: The proposed fusion approach leverages the strengths of ARIMA in modelling temporal dependencies and the ability of LSTMs to capture complex nonlinear relationships. Initially, ARIMA is employed to model the underlying trend and seasonality in the crime data. Subsequently, the residuals obtained from ARIMA are fed into an LSTM architecture, such as LSTM, to capture the remaining nonlinear patterns and dependencies. The hybrid model is trained using historical crime data and validated using appropriate evaluation metrics. The performance of the proposed fusion approach is evaluated on real-world crime datasets across various geographic locations and crime types in India. Results: Experimental results demonstrate that the hybrid ARIMA-LSTM model outperforms individual methodologies and baseline models in terms of MSE, RMSE, and accuracy. Conclusions: This research contributes to the development of more effective crime prediction models, aiding law enforcement agencies in proactive decision-making and resource allocation for crime prevention.