A Comparative Study of Sampling Methods for Imbalanced Credit Card Fraud Detection

Credit card fraud is a major issue worldwide, leading to huge financial losses and undermining consumer confidence in online transactions. One of the biggest challenges in detecting this kind of fraudulent activity is that transaction datasets are usually unbalanced; there are much fewer fraudulent transactions compared to legitimate, valid ones. In addition to this, fraud tactics keep changing, which makes traditional detection systems based on fixed rules less effective. In this paper, we use a machine learning model that uses logistic regression to predict fraudulent credit card transactions. To address the problem of class imbalance, we incorporate a few resampling techniques in the preprocessing steps, these include random oversampling, random under sampling, and cluster centroid sampling. Our main goal is to assess how these techniques affect model training and prediction performance, particularly improving the evaluation metrics such as recall, precision, and overall accuracy. The experiment shows that performing random oversampling improves the model performance significantly as it gives the best balance between precision and recall. This study exhibits how important preprocessing strategies are for improving fraud detection in practical situations. Overall, this research highlights the necessity of using resampling techniques with machine learning models as a way to deal with heavily imbalanced datasets. In the future, more advanced methods of deep learning and reinforcement learning can be taken up to further enhance the fraud detection capability while also making sure it is scalable, protects security and adapts well for real-Time application.