Optimizing Fraud Detection in Credit Card Transactions Using Machine Learning Techniques
Niken Ayu Larasati, Afrig Aminuddin, Hamidah Maulida Khasanah, Kardilah Rohmat Hidayat, Schisin Arsene Douatsap Ndongson, Hendra Dwi Kurniawan · 2025
This research evaluates the effectiveness of machine learning models, including Logistic Regression (LR), Random Forest (RF), Gradient Boosting (GB), and Neural Networks (NN), in detecting fraudulent credit card transactions. Fraud detection is challenging due to the highly imbalanced nature of the data, with fraudulent transactions making up only a tiny fraction of the total. To address this, RandomOverSampler balances the dataset, ensuring a fair representation of both classes. The study employs key evaluation metrics, including accuracy, precision, recall, F1-score, and AUC, to assess model performance comprehensively. Among the models tested, Random Forest demonstrated outstanding results, achieving 99.99 accuracy, 99.99 precision, 100 recall, 99.99 F1-score, and 100 AUC. Gradient Boosting also performed well, with 99.52 accuracy and 99.95 AUC, but RF’s perfect recall and balanced performance across all metrics make it the most reliable model. This research highlights the significance of robust models in practical fraud detection systems, ensuring precise and comprehensive transaction monitoring.