Cost-Sensitive Deep Neural Networks for Financial Fraud Detection

DOI: 10.5176/2251-2136_ICT-BDCS17.31

Authors: Kruy Seng and Man-Leung Wong

Abstract:

Financial fraud has been a serious problem which attracted worldwide attention ranging from finance, economics, and management. Corporate fraud is one of the FBI's highest criminal priorities which caused destructive losses to investors, investor confidence, and country economy. It also contributed to a number of largest bankruptcies in history. In this paper, we propose a cost-sensitive learning approach based on Deep Neural Networks to solve financial fraud detection problem. We introduce cost information into the cost function of Deep Neural Networks to eliminate the bias toward non-fraudulent class. To demonstrate the effectiveness of the proposed method, we conduct comparison experiments with five machine learning and three ensemble learning approaches. One real-life data set of China Corporate Securities Fraud and one benchmark data from UCI repository are used in the experiments. The results show that our method is more capable of detecting fraudulent cases than the other methods, in particular on the real-life data set.

Keywords: fraud; security; neural network; class imbalance;