Credit Card Fraud Detection Using CNN

Credit card is becoming more and more popular in financial transactions,at the same time frauds are also increasing.A large number of credit card fraud bring huge losses to banks and individuals every year.Conventional methods use rule-based expert systems to detect fraud behaviors.In general,the expert systems have fixed structures and are not easy to generalization so as only to detect simple patterns of frauds.Compared to the expert systems,machine learning models are more complicated and have stronger abilities of modeling and generalization.So machine learning methods can detect more fraudulent patterns effectively.More and more scholars are using machine learning methods based on statistics to detect credit card frauds.There are many problems and challenges we need to face in the use of machine learning methods to detect credit card fraud.Firstly,the credit card transaction is a time-series model.How to extract relevant features is an important problem to be solved.In addition,credit card data is extremely imbalanced.The number of fraud samples is far less than the number of normal samples.What kind of method to balance the proportion of positive and negative samples is the key point to improve the efficiency of fraud detection.In this paper,we develop a CNN-based fraud detection framework,to capture the intrinsic patterns of fraud behaviors learned from labeled data.Abundant transaction data is represented by a feature matrix and a convolutional neural network is applied to identify a set of latent patterns for each sample.In the part of feature engineering,we propose a new feature called trading entropy so as to characterize the recent change of customer behaviors efficiently.To solve the problem of imbalanced data,we employ a cost-based sampling method to increase the number of fraud samples.At the same time,in order to make use of the abundant legitimate transaction data,we use the bagging method to train lots of convolutional neural networks and average their results.The fusion of multiple classifiers can significantly improve the robustness of the model.The massive credit card transactions in the experiments are from a commercial bank.The F1-score and ROC curves are used as metrics.The experimental results demonstrate its superior performance compared with some other state-of-art methods.