Flexibility and easiness for making financial transaction has continued to make credit card transaction a popular alternative financial payment especially during Corona-19 Pandemic that hamper almost worldwide. However, recognizing credit card fraudulent transaction is still an open problem despite many methods have been proposed. The main problems, among others, are imbalanced nature of the credit card transaction data and no agreed upon features to represent the transaction. This paper presents a novel approach to address such problem by exploiting card user spending behavior as the main feature to represent the transaction. in our experiment with Random forest we put max features parameter value 18 and 75 as number of trees in Random forest. While our experiment with OCSVM is by using 4 different kernels which are RBF, Polynomial, Linear and Sigmoid, and, we tuned nu and gamma parameter to get the optimal precision, recall and AUC for our Classifier. The experiment results found that One-Class SVM model achieve 0.984 AUC for training and 0.985 AUC for testing; whilst, Random Forest model achieves 0.991 AUC for training and 0.943 AUC for testing.