Efficient fraud detection using deep boosting decision trees

Biao Xu, Yao Wang, Xiuwu Liao, Kaidong Wang

Abstract: Highlights•We propose deep boosting decision tree (DBDT), an effective fraud detection method.•DBDT combines the advantages of both conventional methods and deep learning.•DBDT embeds neural networks into boosting to improve the generalization.•DBDT builds a boosting like end-to-end structure to maintain good interpretability.•We employ a compositional AUC maximization approach to deal with data imbalance.