Geospatial analysis of crime patterns and hotspots for crime prevention using deep learning algorithms

Abstract

ABSTRACT Crime is a major societal challenge that poses a significant threat to public safety and security. Geospatial analysis has proved to be an effective tool in identifying crime patterns and hotspot areas. This work used geospatial analysis for crime prevention by exploring the different techniques, such as Geographic Information Systems (GIS), spatial clustering methods, and hot spot analysis. Objective: To utilize GIS, spatial clustering, and hotspot analysis for mapping crime data, identifying high-risk areas, and uncovering underlying factors contributing to concentrated criminal activity. Methods: The framework involves six stages: data collection, spatial pre-processing, clustering (DBSCAN, KDE), deep learning-based forecasting (CNN, LSTM, and ConvLSTM), model evaluation (RMSE, accuracy, F1-score), and Web-GIS visualization. Crime data from open sources was enriched with spatial features, processed using GIS tools, and used to train and validate forecasting models on spatiotemporal patterns. Results and conclusion: The findings demonstrated that the deep learning-enhanced spatial model effectively captured complex crime dynamics. ConvLSTM achieved the highest accuracy in forecasting future crime incidents, with an F1-score of 0.87, outperforming baseline models. Spatial clustering provided high overlap with known historical hotspots, validating the reliability of unsupervised hotspot detection. The Web-GIS interface enabled intuitive interaction with predictive heatmaps, supporting timely interventions by law enforcement.