ARTIFICIAL NEURAL NETWORKS RESTRICTION FOR ROAD ACCIDENTS SEVERITY CLASSIFICATION IN UNBALANCED DATABASE

• Autor(a) principal: Chuerubim, Maria Lígia
• Data de Publicação: 2019
• Outros Autores: Valejo, Alan, Bezerra, Barbara Stolte [UNESP], da Silva, Irineu
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/249111
• topic: artificial neural networks; classification; road accidents; severity; Unbalanced data

The objective of this study is to discuss the main constraints in classifying the severity of road accidents using Artificial Neural Networks (ANN). To achieve this, ANN modelling with Multiple Layers Perceptron (MPL) was used. This method is recommended for treating non-linear problems, whose distributions are not normal, which is the case for road accidents. Variables associated with the characteristics of accidents, road infrastructure and environmental conditions were used, with the objective of identifying the influence of these factors in the accident severity. The results indicated that ANN modelling with MPL presents a potential association among the parameters related to road accidents. However, the results are limited, since the classification process provides a low rate of accuracy for accidents with victims. Such accidents correspond to less frequent observations in the database, meaning that the data is less represented, and the database becomes unbalanced. Thus, for further research studies, the use of ANN with MPL associated with data balancing methods is suggested, in order to obtain the best data fit to the model and more consistent and realistic results.