Information fusion in crime event analysis: A decade survey on data, features and models

Abstract

Crime event analysis (CEA) has become increasingly important in assisting humans in preventing future crimes. A fundamental challenge in the research community lies in the dynamics of criminal intents. Offenders’ criminal intents may evolve due to their surroundings, making it difficult for machine learning models to capture them from limited existing information, leading to model uncertainty. As a result, there has been a surge of works exploiting various information related to the evolution of criminal intents, thus enhancing prediction accuracy. This work conducts a comprehensive survey of the past decade (2013–2023) of CEA methods from the perspective of information fusion. We first investigate the categories of crime data and briefly introduce existing CEA tasks as well as evaluation metrics. Then, fusion is systematically reviewed from the bases of multi-modal data, features and machine learning models in terms of different categories of crime data. Finally, we conclude by highlighting some limitations and identifying several future research directions.