Crime Type And Occurrence Prediction Using Machine Learning

Abstract

Crime has become a growing concern, disrupting societal balance and posing challenges to public safety. Understanding crime patterns is essential for proactive law enforcement and resource allocation. This study evaluates a machine-learning-based crime prediction system that utilizes open-source crime data to classify and predict recent criminal activities. The system was assessed post-hoc for real-time deployment. Additionally, traditional classification models were compared against a benchmark probability-based classifier (Naïve Bayes) to evaluate classification accuracy. The dataset included crime reports with spatial and temporal attributes, enabling pattern analysis across various crime types. Feature selection and model performance were examined to determine the most influential factors in crime prediction. Results demonstrated that machine learning models, particularly ensemble techniques and neural networks, achieved higher accuracy in crime type classification compared to traditional statistical approaches. Furthermore, crime hotspots and high-risk periods were identified, emphasizing the potential of predictive analytics in crime prevention. The ability to analyze crime trends dynamically is a key advantage of machine-learning models, and their integration into law enforcement strategies can enhance crime prevention and public safety.