DEVELOPMENT OF AN EFFICIENT MACHINE LEARNING MODEL FOR CREDIT CARD FRAUD DETECTION

Credit card fraud poses a significant threat to modern financial systems, resulting in substantial economic losses and
eroding consumer trust. This paper presents an efficient and effective machine learning model for the detection of
credit card fraud by combining multiple classification algorithms along with resampling techniques to minimize the
difficulties faced because of highly imbalanced transactional data that is obtained from a credit card transaction
dataset. The proposed approach integrates Logistic Regression, Decision Tree, K-Nearest Neighbors, Random
Forest, and XGBoost models, each evaluated under various sampling strategies including Random Oversampling,
Random Under sampling, Tomek Links, Cluster Centroids, SMOTE, and SMOTE+Tomek Links to enhance model
robustness and detection accuracy. Performance assessment using precision, recall, F1-score, and AUC-ROC
demonstrates the model’s efficient yet effective ability to differentiate fraud transactions from legal and legitimate
ones. The system also supports dynamic updates through periodic retraining and investigator feedback, which
ensures that the model continuously evolves and adapts to the newly improved and enhanced fraud pattern. In the
future, the team will work to identify, improve and strengthen the real-time decision-making aspect of the machine
learning model so as to improve and further the fraud detection capabilities of the model.