Analysis and Comparison of Credit Card Fraud Detection Using Machine Learning

摘要

Credit card sharp practice detection is one of the most important issues which must be motivated to save the financial institution from huge losses. Several machine laming models such as random forest, logistic regression, Naive Bayes, and XGBoost have been used to verify whether the transaction is fraudulent or genuine one. The data sets which is used in the research has been taken out from European Cardholder consisting of 284,807 transactions. As the data sets are highly imbalanced, so, SMOTE oversampling technique has been used. This experiment is carried out in three phases. First with individual standard model next with soft voting and finally with AdaBoost to know which model gives better results. F1 and MCC have been used for evaluation of the model as accuracy might leads to misclassification problem.