Aplicação de machine learning para predição de crimes no contexto do Distrito Federal

• Ano de defesa: 2025
• Autor(a) principal: Vaguetti, Hially Rabelo
• Orientador(a): Paiva, Matheus Silva de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Católica de Brasília
• Programa de Pós-Graduação: Programa Stricto Sensu em Governança, Tecnologia e Inovação
• Departamento: Escola de Educação, Tecnologia e Comunicação
• País: Brasil
• Palavras-chave em Português: Políticas de segurança; Criminalidade; Dados socioeconômicos; Inteligência artificial; Predição de crimes
• Palavras-chave em Inglês: Security policies; Crime; Socioeconomic data; Artificial intelligence; Crime prediction
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://bdtd.ucb.br:8443/jspui/handle/tede/3693

Crime is a complex phenomenon influenced by socioeconomic factors, requiring innovative approaches for its analysis and prediction. This study investigates the effectiveness of machine learning models in crime prediction in the Federal District, considering crime and socioeconomic data obtained from official agencies, such as the Federal District Planning Company and the Federal District Public Security Secretariat. The Random Forest and Extreme Gradient Boost models were applied, focusing on selecting the most relevant variables, evaluating model effectiveness, and analyzing the impact of class balancing techniques on the constructed models. The results indicate that while positive predictions are possible, the models exhibited limitations, particularly due to the high rate of incorrect predictions and challenges in class selection and balancing. Despite these constraints, the application of machine learning proved to be a valuable tool for predictive crime analysis, provided that methodological refinements and improvements in data quality are implemented. This study highlights the importance of integrating AI techniques into public security policymaking. Finally, it suggests the need for further research to optimize available data and enhance its applicability in crime prediction within the Federal District.