Modelling of the Customers' Propensity to Fraud in Automobile Insurance

This thesis is devoted to the problem of fraud in insurance. The paper was carried out to investigate the issue of fraudulent claims in Motor Insurance. As for the Russian market such studies had not previously been conducted, the analysis was to reveal some peculiarities for Russia. The purpose of this work was not so much to provide a model that could be applied in insurance companies to identify fraud in insurance, as much to present the approach of using data mining techniques to identify suspicious claims and to test the applicability of the use of scoring techniques for these purposes, and to identify relationships that are unique for the Russian market. On the basis of the inspected literature the technique to use for building a model was defined – scoring model based on logistic regression, the factors with strong positive impact on fraud possibility were stipulated, the hypothesis about the applicability of this model to solve the problem of the study and the significance of popular and specific factors were put forward. In the course of the study to test the hypotheses, data was collected and processed, containing information about the claims, consisting of 1339 observations, 337 of which are fraudulent. The model was built based on it, witch with a fairly good accuracy determines the chance of fraud. The results were obtained the following: it was confirmed strong positive effect of fraud popular factors as well as of identified specific factors of the Russian market on the likelihood: claim registration channel and belonging to a region with a high level of risk and loss exposures.