Variable Selection and Asymmetric Links to Predict Credit Card Fraud

Credit card fraud identification is a challenging problem for different reasons: it needs to be suddenly detected; it is based on the use of huge data sets that have to be properly managed; the number of fraudulent transactions is definitely lower than the number of genuine transactions and then, this imbalance requires the use of proper statistical models. Here we discuss how the data reduction, performed through the variable selection, can be combined with the use of Generalized Linear Models with asymmetric link functions which are able to handle imbalanced data. We illustrate how these theoretical results can be used for credit card fraud-detection purposes.