Crime prediction before during and after COVID 19 using machine learning and RNN LSTM models

Abstract

One of society’s foremost challenges is crime, and over the years, statistical methods have been employed for crime prediction, albeit with limited success rates. Recently, the utilization of machine learning and deep learning algorithms has significantly improved the precision of crime forecasting. However, existing studies mostly focus on crime prediction but often fail to meet practical policing requirements and the impact of the COVID-19 pandemic. The Lack of in-depth predictive capabilities motivates the need for a framework that incorporates police patrol planning by considering crime trends across the pre-pandemic, pandemic, and post-pandemic periods. Several machine learning algorithms, such as logistic regression, decision trees, random forests, XGBoost, and the deep learning algorithm RNN-LSTM, were used in this research. The exploratory analysis is additionally added to acquire insights into the pattern of crime before, during, and after the COVID-19 pandemic and what factors are influencing this pattern. These models are evaluated by comparing their accuracy of these models. The outcome of this research will give early warning of criminal activity, highlight high crime areas where the rate of crime is higher, and predict future trends with greater accuracy than existing prediction methods. These insights will provide value by informing law enforcement practices and strategies. The study used 2,124,602 crime records from the Chicago crime dataset spanning 2015–2023. Among the machine learning models, the optimized XGBoost classifier achieved the highest accuracy of 91.12%, while the RNN-LSTM model delivered the best overall performance with an accuracy of 92.74%. The COVID-19 analysis further revealed noticeable variations in crime patterns and crime intensity across different districts of Chicago during the pandemic period.