Internet money has lately becoming somewhat trendy. Still, bad debt now poses a major danger to Internet financial firms. Usually utilized in traditional financial firms, logistic regression is the fraud detection model. While the logistic regression is interpretable, its accuracy still has to be raised. This article investigates the possibilities of using deep neural networks for fraud detection using a big public loan dataset, like Lending club. Using a random forest, we first address the missing data. The most discriminatory characteristics are then chosen using an XG Boost technique. To handle the sample imbalance after that, we suggest to apply a synthetic minority oversampling approach. We build a deep neural network using preprocessed data to detect Internet loan fraud. To show the out performance of the deep neural network against the widely used models, many tests have been carried out. Such a basic but powerful model might improve the use of deep learning in anti-fraud for Internet loans, so benefiting the financial engineers in small and medium Internet financial firms.

Internet loan, anti-fraud, guardiannet, neural network.