Data science aplicada à análise criminal baseada nos dados abertos governamentais do Brasil

• Autor(a) principal: Prado, Kleber Henrique de Jesus
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFS
• Texto Completo: https://ri.ufs.br/jspui/handle/riufs/14190
• topic: Computação; Análise criminal; Banco de dados; Administração pública; Ciência de dados; Association rules; Criminal analysis; Data science; Meta-analysis; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Context: Crime is a common and complex social problem that affects a nation’s quality of life, economic growth and reputation. Governments and society in general have had enormous problems caused by this phenomenon. Each year, governments spend millions of dollars fighting violence and, consequently, crime prevention and control are issues of great concern to public security agencies. Therefore, new approaches and advanced systems are needed to improve crime analysis and to protect their communities. In this context, Data Science has been playing a vital role in improving the results of criminal investigations and detections, facilitating registration, recovery analysis and information sharing. Objective: Apply fundamentals of Data Science and provide an automated model, constantly updated, to analyze open government data related to crimes occurred in the Federative Units (FUs) and in the municipalities of Minas Gerais. Method: Initially, we performed a quantitative Systematic Review (SR) (with meta-analysis), as a way to identify and systematize the main approaches, techniques and algorithms used in the intelligent analysis of open government data related to criminal initiates. Then, we performed two experiments to discover rules of association between states, municipalities, crimes, Integrated Public Security Regions (IPSRs), theft targets and theft targets. Additionally, we detect outliers in relation to crime rates and developed rankings that show the most dangerous locations (states, municipalities or IPSRs). Results: In the context of Brazilian states, from a general point of view, with weights for crimes, Paraná was the most dangerous place in all the years evaluated. Also noteworthy for Rio de Janeiro, always occupying the second position. Besides that, the states of Goiás, Pernambuco and Rondônia were classified among the five most dangerous, in three of the five years analyzed. From the single perspective of murders, in 2019, the states of Roraima, Rio Grande do Norte, Sergipe, Acre and Pernambuco were ranked as the ten most violent ones, being Pernambuco and Acre among the most dangerous states from the two perspectives (weighted average and murders). Regarding the association rules, it became evident that there are dependencies between crimes and states. In the context of the state of Minas Gerais, Belo Horizonte, Confins and Contagem were constantly among the five most dangerous municipalities. In addition, it became evident that there are dependencies between: crimes and municipalities, crimes and IPSRs, robbery targets and municipalities, and robbery targets and IPSRs. However, no associations were detected between theft targets and municipalities, and theft targets and IPSRs. Conclusion: Data Science enables the execution of more accurate and faster diagnostics, helping strategic planning and decision making in Public Security.