Predictive Crime Analysis Using Multi-Layer Perceptron Architecture

Abstract

In an extended period, crime and statistical professionals’ analyses have channeled their skills, knowledge, and expertise to anticipate the timing and locations of future criminal incidents, although with varying degrees of success. The surge in criminal activities and the evolving strategies adopted by modern offenders have strained the efficacy of existing predictive methods. This study introduces a novel approach by leveraging the Multi-Layer Perceptron (MLP) architecture, a cutting-edge technology that uses the back-propagation algorithm to develop a predictive model for analyzing crime data. A total of 4,748 records were collected from the Cross River State Police Command. Data training was conducted using MLP, and the dataset was divided into 70% for training and 30% for testing. The outcomes of the MLP model, characterized by a precision of 0.84, an accuracy of 74%, a recall rate of 0.73, and an F1-score of 0.79, underline the suitability and effectiveness of employing MLP as an invaluable tool in crime prediction.