Aplicação de Machine Learning em seguros de autos

Abstract

The present work aims to price the insurance premium value of a vehicle protection association using the Machine Learning approach through the Random Forest algorithms, Bootstrap aggregating with the application of the Random Forest regression. The main metrics for evaluating the results of the application of regression by Random Forest were RSME, MAE and graphical analysis, data transformation techniques and Principal Component Analysis, PCA were also used. Two response variables were used for two different models, the first was the variable referring to the occurrence of claims, which results in the average importance of the predictor variables regarding the frequency of claims and the second was the variable referring to the indemnity of claims, its result showing average importance predictor variables as to the severity of claims. The expected loss ratio pricing method was used to obtain the value of the collective risk premium and for the collective pure premium, the collective pure premium was used as a parameter to calculate the annual individual pure premium, the annual individual premium was calculated for each risk from the importance of the risk variables obtained by the Random Forest regression method and having these values, we calculated the individual annual premium for different profiles with the same value of the insured amount, thus, it was possible to make an evaluation of the applied method.