Credit Card Fraud Detection System using SMOTEENN and Adaptive XGBoost and comparing the result with state-of-art-technique

The number of online transactions has increased over the years. India alone is said to have made 8 billion digital transactions last year, and is only increasing every year. With the increase in online transaction, fraudulent transaction has also increased, which has cost financial institutions in billions of dollars. Therefore, there is a need to stop such fraudulent transactions and various researchers have proposed machine learning models to predict such fraudulent transactions. However, datasets belonging to credit card fraud detection are highly imbalanced as it contains non-fraudulent transactions as majority class as in the ratio of 99:1 when compared to fraudulent transaction. This makes any machine learning model to be biased towards the negative class. In this paper, we have performed machine learning models on two datasets. One imbalanced data-set, and the other as balanced data-set using SMOTEENN. We performed Adaptive XGBoost on the two datasets and measured its performance using the metrices of Accuracy, Precision, Recall, Specificity, FPR, AUC, MCC, G-mean and the time taken to run the model. The result was compared with various state-of-art machine learning models such as Logistic Regression, KNN, Decision Tree, GaussianNB, AdaBoost Classifier, and Random Forest Classifier.