Data-driven intelligence for urban crime analysis and perception

Abstract

Brazilian agencies in charge of public safety lack state-of-the-art tools to extract knowledge from the massive datasets. The few city security agencies with access to up-to-date technologies employ those technologies mainly for surveillance, not accomplishing complex analytical tasks to plan new public policies against crimes. The technology used in those few Brazilian cities only detects real-time crime situations rather than extract knowledge from massive data to drive general guidelines on crime prevention. It is worth mentioning that surveillance, patrolling, and repression are policies long used by the police with favorable results in terms of crime reduction but with little impact on the insecurity feelings of the population. This proposal aims to create a machine learning and data visualization framework to explore big data already collected by security agencies. We will create tools that serve government agencies to make decisions and the general population be informed (since citizens need to know the insecurity level of the region where they live). This project has three mains’ goals: 1) Acquire socioeconomic, infrastructure, and crime appearance data and estimate their frequency over different city districts. 2) Study their relationship to predict new crimes and estimate security perception using three strategies: i) use traditional statistical and machine learning tools to explore relations among the variables; ii) explore graph signal processing and deep learning to extract temporal patterns and forecast crime occurrences; iii) use deep learning to capture security perception on street images. 3) Create visual tools to help experts to understand the patterns. As a result, we expect to create a helpful tool that would help governments and public security agencies invest in new analytical solutions for crime prevention in such a decade where Brazilian economic panorama is impairing investment from most city governments. (AU)