Multi-UAV Allocation Framework for Predictive Crime Deterrence and Data Acquisition

The recent decline in the number of police and security force personnel has raised a serious security issue that could lead to reduced public safety and delayed response to crimes in urban areas. This may be alleviated in part by utilizing micro or small unmanned aerial vehicles (UAVs) and their high-mobility on-board sensors in conjunction with machine-learning techniques such as neural networks to offer better performance in predicting times and places that are high-risk and deterring crimes. The key to the success of such operation lies in the suitable placement of UAVs. This paper proposes a multi-UAV allocation framework for predictive crime deterrence and data acquisition that consists of the overarching methodology, a problem formulation, and an allocation method that work with a prediction model using a machine learning approach. In contrast to previous studies, our framework provides the most effective arrangement of UAVs for maximizing the chance to apprehend offenders whilst also acquiring data that will help improve the performance of subsequent crime prediction. This paper presents the system architecture assumed in this study, followed by a detailed description of the methodology, the formulation of the problem, and the UAV allocation method of the proposed framework. Our framework is tested using a real-world crime dataset to evaluate its performance with respect to the expected number of crimes deterred by the UAV patrol. Furthermore, to address the engineering practice of the proposed framework, we discuss the feasibility of the simulated deployment scenario in terms of energy consumption and the relationship between data analysis and crime prediction.