Luis Alexandre Rodrigues and Nizam Omar, Mackenzie Presbiterian University, Brazil

Through a cost matrix and a combination of classifiers, this work identifies the most economical model to perform the detection of suspected cases of fraud in a dataset of automobile claims. The experiments performed by this work show that working more deeply in sampled data in the training phase and test phase of each classifier is possible obtain a more economic model than other model presented in the literature.

Fraud Detection, Multi Classifier, Data Mining.