Classification model applied to fraud detection within the insurance sector

Abstract

Fraud is a worrying activity that affects several factors as a whole. With the objective of addressing this issue adequately, it is necessary to build methods that detect and prevent this type of activity. In this project, we propose a text analysis tool for the detection of fraudulent behavior and two models for its prediction, and where we have made use of Zurich Insurance PLC data of 2020 and 2021 to conduct this study. The results that have been obtained demonstrate that there is a lot of variability in the data, where the prediction of fraud through the models has obtained an accuracy of around 63%. Nevertheless, the study that has been carried out has allowed us to extract some relevant conclusions, together with some considerations for future research.