Data mining and the quality of extracted knowledge from police reports of Brazilian Federal Highways

• Autor(a) principal: Costa, Jefferson de Jesus
• Data de Publicação: 2014
• Outros Autores: Bernardini, Flávia Cristina, Viterbo Filho, José
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: AtoZ (Curitiba)
• Texto Completo: https://revistas.ufpr.br/atoz/article/view/41346
• topic: Computer Science: computing methodologies and techniques; Open Government Data; Data Mining; Association Rules; Knowledge Discovery in Databases; Ciencias de la Computación: metodologías y técnicas de computación; Datos abiertos gubernamentales; Minería de datos; Reglas de asociación; Descubrimiento de Conocimiento en Bases de Datos; Ciência da Computação: metodologias e técnicas da computação; Dados Governamentais Abertos; Mineração de Dados; Regras de Associação; Descoberta de Conhecimento em Bases de Dados

Introduction: This paper presents and analyzes the results obtained when applying Data Mining process in the bulletins of occurrences of the Brazilian federal highways generated by the Federal Highway Police (PRF) in 2012. The purpose of this work is to analyze the feasibility of implementing the Data Mining process on data provided by PRF in order to identify associations between variables related to transit accidents in all Brazilian federal highways. Method: It was used symbolic supervised learning algorithms, as well as an algorithm of generation of association rules, implemented in Weka tool. Regarding the database, it was used the records of 2012. On this portion of the database it was conducted the step of data preprocessing, which were used for extracting models and patterns in the Weka tool and, lastly, evaluated the models and extracted patterns. Results: In supervised learning, the results obtained with J48 and PART algorithms have been considered promising due to the fact that for all classes of accidents causes, the values of area under the ROC curve (AUC) were above 0.5. Furthermore, using the Apriori algorithm there have been generated 38 association rules with confidence greater than 0.8. Conclusions: It was concluded that is important to propose a model for data distribution of this database, in order to use it for data mining process, as well as other knowledge extraction tasks and decision making. It was noted still, the need to improve the quality of data to be provided from the initial stage of data gathering, that is, in the very systems used to record the data.