Vehicle claims in the south of Minas Gerais: an approach using classification models

Abstract

With the changes in the patterns of risk, new insurance products are available on the market. Consequently, pricing models are restructured to manage levels of risk and create premiums that maintain the well-being of insurers. This work analyzed the Logistics and Random forests models in the classification of total loss events in the south of Minas Gerais using original and artificial samples, built by the ROSE resampling method, which is a procedure for constructing artificial samples in a smoothing bootstrap. A total loss of a vehicle is considered when the repair costs for the same event exceed a percentage established by contract. As a result, it was obtained that the models with artificial data improved the balanced accuracy rate on unbalanced data.