Crime Prediction Linked to Geographical Location with Periodic Features for Societal Security

To understand the future relationship between crime locations and crime types, traditional security systems use historical crime records. These records are manually analyzed after incidents with the knowledge of domain experts. Manual analysis of crime data in real-world settings is prone to misinterpretation and prediction error. Recent studies have investigated crime-related problems with static features of crime data that hinder the correct predictions of security systems. Therefore, in this study, a machine learning system based on derived periodic features and an ensemble model is developed to solve the difficult problem of security systems. Experiments were conducted using data sets from the American police department. The proposed model outperforms conventional approaches in predicting the likely associations between crime types and locations, with an average overall precision of 0.97, recall of 0.96, F1 score of 0.96, and accuracy of 96.61. The deployment of this proposed model in a complex environment can help security personnel to have foreknowledge of the crime before it occurs, as well as in the decision-making process.