A hyper-parameter tuned Random Forest algorithm-based on Artificial Bee Colony for improving accuracy, precision and interpretability of crime prediction

Abstract

Crime prediction plays a crucial role in enhancing public safety and optimizing resource allocation for law enforcement. Traditional methods often fall short in addressing the complex and dynamic nature of crime data, relying on oversimplified assumptions and limited datasets that reduce accuracy and effectiveness. Advanced machine learning techniques, particularly a hyper-parameter tuned Random Forest model optimized using Artificial Bee Colony (ABC) algorithms, present a promising solution. This study proposes an enhanced crime prediction methodology that incorporates ABC-based hyperparameter tuning and Recursive Feature Elimination with Cross-Validation (RFECV) to improve accuracy, interpretability, and robustness. The model leverages ensemble techniques to integrate diverse features from historical crime data, capturing intricate crime patterns more effectively. Performance evaluations will compare the proposed approach with existing models using metrics such as Predictive Accuracy Index (PAI), Predictive Efficiency Index (PEI), Recapture Rate Index (RRI), and SHapley Additive exPlanations (SHAP) values. By prioritizing accuracy, transparency, and stakeholder engagement, this research aims to develop reliable, interpretable, and data-driven crime prediction models, fostering informed decision-making and proactive crime prevention.The work emphasizes improving crime prediction models through advanced machine learning techniques, including enhanced model development, integration of diverse data sources, focus on interpretability, continuous optimization, and stakeholder engagement. These recommendations aim to create robust, interpretable, and data-driven models that support law enforcement decision-making while addressing biases and existing limitations.