MAPFRE Insurance

Solution

MAPFRE Insurance built an in-house fraud-detection workflow around Conditional Tabular Generative Adversarial Networks (CTGAN) for its US insurance operations. The process starts with historical claims data, where fraudulent cases are relatively rare and therefore hard to use as a balanced training set. MAPFRE uses CTGAN to learn the statistical structure of that tabular data and generate synthetic claims records that preserve privacy while expanding the fraud training population. Those synthetic datasets feed AI systems that combine machine learning and graph analysis to identify fraud patterns across claims, first in motor insurance and later in home insurance. In production, when a home insurance claim is filed, the trained model estimates the likelihood of fraud and flags suspicious cases for further investigation by claims teams. MAPFRE presents the approach as a privacy-conscious way to improve model precision, streamline claims processing, and reduce fraud-related losses.

Data flow

Historical motor and home insurance claims data, including limited fraud examples, are used to train a CTGAN model that generates synthetic tabular records. MAPFRE Insurance uses the resulting balanced dataset to train machine-learning and graph-analysis fraud models. When a home insurance claim is submitted, the trained model scores its fraud likelihood and flags suspicious cases for further investigation.