A Synergistic Ensemble Approach for Enhanced Time Series Forecasting in Crime Analytics

This study proposes a novel ensemble approach of RF Regression, ANNs, and ARIMA for time series forecasting. To address the objectives of the present research, sensitive and restricted crime data obtained from the Mumbai police has been used. It covers a number of years and entails detailed records of various categories of offenses that have been committed in the area monthly. Several techniques of data preprocessing are applied to ensure that the data in the given set is fit to train the model and they include data cleaning, scaling, and handling of missing values. The hybrid ANN-ARIMA-RF approach captures the special advantages of each component technology: ANNs are employed for nonlinear patterns while ARIMA is employed for the linear patterns in residuals and RF is applied on both models outputs. The hybrid of ARIMA and ANN outperforms the individual ARIMA and ANN models when applied to the exclusive crime dataset to forecast future crime rates. This is because the developed model improves the prediction accuracy as shown by the lowest Mean Absolute Percentage Error. This study differs from other works due to the utilization of a unique dataset and the integration of several forecasting methods that improve upon the individual ones.