Construcción de algoritmo para análisis de causa raíz de accidentes de tránsito utilizando redes neuronales y minería de datos.
Fecha
Autores
Autor corporativo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Compartir
Director
Altmetric
Resumen
Determining the root cause of transit accidents is crucial to understand events dynamic. There are a great number of techniques to do this. More of them implies the participation of stakeholders in transit accidents analysis, using their experience. Data mining is an emergent technology that can be used in the analysis and solution of different complexity problems, like traffic accidents. It allows experts to extract knowledge from the information in the data gathered in the accident site. In this work it has been employed two databases; the first contains 38 fields and 34628 registers and is available in (https://www.datos.gov.co/widgets/79fi-zm8c. The second one supplies information taken from Allianz traffic accidents files and includes only information of transportation companies. It has and 890 registers Root cause model has been divided in two sections; one for find the immediate cause and the other for search the possible root cause of the accidents. To do this it has been proposed three different algorithms: Multilayer Perceptron (MLP) and Naïve Bayes(NB) for the immediate cause and PART for the basic or root cause. Performing the data mining utilizes the free software WEKA 3.8. le evaluation of the performance of that algorithms is based on the use of metrics like percentage of hits and misses, kappa index, root mean square error, ROC, Concerning the immediate cause of the accidents for the first databe It has been found that Naïve Bayes performs better than ANN multiulayer perceptron, getting 76,44% of correctly classified instances while perceptron does 54,4%. For the second database, MLP got 72% of correctly classified instances. Finally, the models constructed from the algorithms related, was tested in a real case corresponding to a road a truck accident .