Estudo de caso com abordagem das redes Bayesianas no ramo de seguros automotivo

Abstract

Decision making in the face of partial or inaccurate information that needs techniques to deal with uncertainties suggests the use of Bayesian Networks, which have relatively well-disseminated studies in various areas of human knowledge, but still little exploring the insurance area. There are various types of insurance on the market and the current population demands technology, new products that reduce risk and coverage according to their needs. The most common are auto, property, life, personal accident and health insurance, among others. Insurance is a very important protection service for the car and many people still avoid it, most often related to its price. The present work aims to verify the degree of knowledge and importance of motor vehicle insurance in a population of students of the exact courses of the Federal University of Sergipe (UFS) and will apply in a pioneering way the techniques of Bayesian Networks for analysis. For the acquisition of the data a pilot project was elaborated using stratified sampling, applying a questionnaire to the students of the Center of Exact Sciences and Technology (CCET) of UFS. These data were explored with the aid of the R Project software with an emphasis on the association packages between variables and the formation of the structure of the Bayesian Networks. Using the association rules mining technique and scoring algorithms, the network initially generated did not include the variables importance and knowledge about automotive insurance, even noting that 91% of the total respondents rated between 7 and 10 for the importance of insurance. These variables were entered manually resulting in a network with associations contrary to the conditional reality of each variable. Thus, despite the changes and meeting the objectives, the network was able to express the importance of vehicle insurance related to the automobile industry through the income variable, even with a median knowledge among the interviewees, through the obtained probabilities.