ENHANCING PUBLIC SAFETY: CRIME CATEGORY PREDICTION THROUGH FUSION-BASED ENSEMBLE LEARNING WITH XG-BOOST

Abstract

The rate of crime has been rising on daily basis all around the world, leading to an increase in security concerns, Closed-Circuit Television (CCTV) cameras have been deployed with the intention of reducing crime and enhancing public protection. The correctness of criminal detection is greatly increased by the use of CCTV cameras. However, the amount of data produced every day by modern monitoring technologies makes it challenging for experts in computer vision to evaluate, since it requires a lot of time and human involvement. The development of a system that can recognize the patterns of human by monitor humans in real-time is necessary.  Decision trees are an effective supervised learning strategy for categorizing and comprehending crime trends because they divide data into essential qualities such as time, location, and victim information, resulting in a visible, rule-based framework. This interpretability makes them useful for detecting influential aspects in crime, but its simplicity restricts usefulness in practical criminology and public safety. While individual classifiers such as Decision Trees, Random Forests, or Support Vector Machines constitute the basis of crime prediction research, they frequently fail to capture the intricate spatiotemporal and categorical correlations in crime datasets. To address these limitations, this study provides a hybrid fusion modeling framework that integrates numerous classifiers using approaches like soft voting to improve forecast accuracy and robustness. The use of advanced algorithms like XG Boost, as well as feature engineering strategies and ways for regulating class imbalance, assures better performance in detecting and predicting criminal patterns. This approach not only takes advantage of the characteristics of several classifiers, but it also offers a scalable and dependable solution for real-world crime analysis.