FORECASTING TRANSACTION CARD FRAUD USING BOOSTING ALGORITHMS

ABSTRACT

Cashless transaction is now becoming famous by the day, as there are various means of making payment aside from carrying cash. This study proposed ensemble learners – adaptive boosting (AdaBoost) and extreme gradient boost (XGBoost) algorithms for predictive model. The data used for this study was collected from a secondary source, Kaggle, the data collected was an imbalanced data, it was statistically analyzed using correlation matrix. The data was used to trained and test the model, using Adaboost and XGBoost algorithms, with python-3 simulation environment. For validation of the models, evaluation metric used include confusion matrix, ROC curve, ROC and AUC score, recall, f1-score, etc. From the result obtained from the model, both boosting algorithms adopted show high level of ROC-AUC score, with Adaboost algorithm slightly outperformed the XGBoost algorithm. The study concluded that both algorithms are suitable for accurately predicting fraud associated with transactions and it is recommended that they can be incorporated into fraud detection system.

Keywords: Forecasting, Transaction Card Fraud, Boosting, Machine Learning