A novel approach based on XGBoost classifier and Bayesian optimization for credit card fraud detection
Nowadays, detecting fraudulent transactions has become increasingly important due to the rise of online businesses and the increasing use of sophisticated techniques by fraudsters to make fraudulent transactions appear similar to genuine ones. Researchers have explored a lot of machine learning clas...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
KeAi Communications Co., Ltd.
2025-12-01
|
| Series: | Cyber Security and Applications |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772918425000104 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Nowadays, detecting fraudulent transactions has become increasingly important due to the rise of online businesses and the increasing use of sophisticated techniques by fraudsters to make fraudulent transactions appear similar to genuine ones. Researchers have explored a lot of machine learning classifiers, such as random forest, decision tree, support vector machine, logistic regression, artificial neural network, and recurrent neural network, to secure these systems. This study proposes an enhanced XGBoost algorithm for detecting fraudulent transactions using an intelligent technique that tunes the hyperparameters of the algorithm through Bayesian optimization. To test the performance of our solution, several experiments are conducted on two credit card datasets consisting of both legitimate and fraudulent transactions. To prevent overfitting on imbalanced datasets, we employed cross-validation, SMOTE, and Random under-sampling techniques. For Data 1, the best performance using SMOTE achieved an accuracy of 0.9996, precision of 0.9406, recall of 0.8740, F-measure of 0.8740, and AUC of 0.9879. For Data 2, the Random Under-sampling technique yielded the highest performance with an accuracy of 0.8325, precision of 0.8294, recall of 0.8378, F-measure of 0.8336, and AUC of 0.9088. Our proposed solution outperforms other machine learning models, as demonstrated by these experimental results. |
|---|---|
| ISSN: | 2772-9184 |