本研究使用 Kaggle 競賽平台上的歐洲信用卡交易資料進行實驗，以評估所提出方法之效能，並與相關的方法進行比較。實驗結果顯示，在準確率(accuracy)、真陰率(true negative rate)、真陽率(true positive rate)和馬修斯相關係數(Matthews correlation coefficient)評分標準下，本論文所提方法皆有較佳的效能。;With the innovation of e-commerce technology and the popularization of mobile devices, shopping has become diverse and convenient, which has prompted more and more people to join the ranks of online shopping, and has also driven the people’s demand for credit cards. However, many crimes are hidden behind the convenience of credit cards. With the gradual increase in the volume of credit card transactions, fraudulent transactions have become more rampant, which has brought huge losses to banks and merchants. Therefore, card issuers hope to establish an effective method for detecting fraudulent transactions of credit cards.
This thesis proposes a credit card fraud detection method. The proposed method first utilizes adaptive synthetic (ADASYN) sampling to oversample the minority class, increasing the number of samples that are harder to learn. Then it uses the DeepInsight method to transform non-image data into well-organized images, which in turn are fed into deep learning convolutional neural network (CNN) model to extract critical features hidden in the raw data for improving the accuracy on credit card fraud detection.
This study uses the European credit card transaction data on the Kaggle competition platform to evaluate the effectiveness of the proposed method and compares the evaluation results with those of related methods. The comparisons show that the proposed method has comparably good performance in terms of the accuracy, true positive rate, true negative rate, and Matthews correlation coefficient.
