Geração de mapas de hotspots em redes de ruas para predição de crimes

Detalhes bibliográficos
Ano de defesa: 2020
Autor(a) principal: Nunes Junior, Francisco Carlos Freire
Orientador(a): Não Informado pela instituição
Banca de defesa: Não Informado pela instituição
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Não Informado pela instituição
Programa de Pós-Graduação: Não Informado pela instituição
Departamento: Não Informado pela instituição
País: Não Informado pela instituição
Palavras-chave em Português:
KDE
Link de acesso: http://www.repositorio.ufc.br/handle/riufc/51515
Resumo: Crimes (e.g., assault, arson, harassment, and murder) have emerged as one of the most critical problems countries face. In particular, in Brazil, crime is a theme of growing interest and the prime concern in some cities, due to the high crime rates, the sheer magnitude of violence and the perceived number of lives lost. A tool created with the use of technology to help tackle crime is the construction of hotspots maps, which are geographically limited regions and have a high concentration of crimes according to historical data. A relevant amount of approaches available in the literature address this problem by suggesting that Kernel Density Estimation (KDE) can accurately forecast crime and outperform other approaches for crime prediction. However, none of these approaches approximate the crime hotspots to the road network by considering that the police patrols move constrained by road networks. In this perspective, this work proposes the creation of four new techniques for generating hotspots maps: Polygon Hotspots Approximated to Road network (PHAR), Incremental Polygon Hotspots Approximated to Road network (i-PHAR), Subgraph Hotspots Approximated to Road Network (SHAR), and Expansive Network, that use KDE density estimates to create hotspots approximated to the streets, with the aim of predicting new occurrences of crimes. We conduct several experiments using real data of theft crimes from Fortaleza, Ceará, Brazil, that demonstrate the PHAR and i-PHAR techniques present results close to KDE algorithm using grid cells concerning the prediction of future events. Moreover, both techniques create fewer hotspots than the baseline algorithm for the same parameter settings. For what concerns SHAR and Expansive Network techniques that create hotspots as subgraphs (of the road network) facilitating patrol planning. SHAR yields superior results in terms of usability and Expansive Network better prediction than the results from KDE algorithm using grid cells.