A Credit Card Fraud Detection Model Based On CNN-SVM

With the rapid development of science and technology,credit card not only provides great convenience to our life,but also brings great growth to economy.At the same time,there are more and more violations and frauds in the process of credit card transactions,which have brought huge economic losses to the country and the society.Here,in order to reduce the losses of banks and cardholders,many researchers are working on accurate and efficient fraud detection systems.Statistical data show that credit card transaction data have problems such as large transaction volume,many characteristics and high imbalance(the normal sample size is much higher than the fraud sample size).It is difficult for the fraud detection system to classify the transaction data set due to the extremely unbalanced data.At the same time,the problem of dimensionality disaster exists in the transaction data set.When the sample data set is high-dimensional data,the performance of the algorithm is poor.This paper proposes a credit card fraud detection method and system based on CNN-SVM.In order to improve the classification accuracy,the SK-SMOTE algorithm is proposed based on the traditional over-sampling method(SMOTE)algorithm.SK-SMOTE can remove noise points or outliers to make the collection samples more concentrated.For the generated data,the SK-SMOTE algorithm greatly improves the randomness of the data range.It makes the model more accurate and reasonable by using the new data from fitting.This paper first uses SK-SMOTE algorithm to deal with the imbalance problem of original data,then carries out effective implicit feature extraction through CNN,and finally makes use of SVM's advantage of accurate data classification to carry out classification detection on the data after feature extraction.The performance of the fraud detection system is evaluated by using the public credit card transaction data set of European banks.The experimental results show that the accuracy and F1 value of the fraud detection system based on CNN-SVM are slightly higher than the CNN algorithm in the detection of positive samples(fraud samples),and far higher than other statistical learning algorithms.The AUC of this detection system reaches 95%,which is about 10% higher than the performance of the traditional fraud detection system,which proves the superiority of this system.