Crime Prediction Through Collaborative Analysis of Proximate Police Stations Data

Abstract

Crime prediction is a crucial aspect of law enforcement strategies and crime prevention efforts. Machine learning has emerged as a valuable tool in crime prediction, allowing for more accurate and data-driven forecasting. In this study, we focus on forecasting the number of crimes at Pathumwan Police Station in Thailand. Utilizing criminal records from various police stations across Thailand, we employ the K-Means clustering algorithm to group police stations exhibiting similar crime patterns to the Pathumwan Police Station. The clustering results reveal that Wang Thonglang, Nang Loeng, Dusit, Bang Sue, Thung Maha Mek, Samre, Sutthisan, Pak Khlong San, Bangkok Yai, Bangkok Noi, Makkasan, Bang Yi Ruea, and Talat Phlu Police Station are clustered together with Pathumwan Police Station. Subsequently, we apply the Long Short-Term Memory (LSTM) model to forecast crimes at Pathumwan Police Station. The training dataset comprises paired data from police stations within the same cluster as Pathumwan Police Station. Our findings indicate that combining data from Wang Thonglang, Nang Loeng, Dusit, Bang Sue, Thung Maha Mek, Samre, Sutthisan, Pak Khlong San, Bangkok Yai, Bangkok Noi, Makkasan, Bang Yi Ruea, and Talat Phlu Police Station with Pathumwan Police Station results in lower errors in Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) compared to using only Pathumwan Police Station data. This collaborative approach enhances the accuracy of crime prediction models and contributes to more effective law enforcement strategies.