A combined approach of the supervised autoencoderand XGBoost method for credit card fraud detection

Losses related to fraudulent transactions areincreasing, so building a fraud detection system is essential.Previous studies have employed a variety of data mining andmachine learning techniques to construct fraud detectionsystems. This study presents a new hybrid method based on thesupervised autoencoder and the extreme gradient boosting(XGBoost) method. This combined method uses the power of asupervised autoencoder to generate an expressive representationof the data. It employs the XGBoost method as a robust classifierto detect fraudulent transactions. The hyperparameters of theproposed method are fine-tuned using the Bayesian optimizationalgorithm. The experiments on a public dataset containing ۲۸۰thousand records demonstrated that the proposed methodachieves better results than the baseline method considering allthe performance criteria, including Recall, Precision, and F۱measure.

Keywords:

Fraud detection , Representation learning , Deeplearning , Extreme gradient boosting , Classification