UCLA

Existing credit card fraud detection studies are generally based on imbalanced datasets,where various sampling methods are used to alleviate the imbalance before applying machine learning techniques for detection. The aim of this study is to apply machine learning methods to a balanced dataset in order to evaluate the reliability of these methods when used on imbalanced datasets. We used a category-balanced data set from Kaggle, which includes 568,630 credit card transaction records. Through an in-depth analysis of extensive real-world transaction data, the research evaluated the performance of algorithms such as Random Forest, Naive Bayes, Backpropagation Neural Networks, and Logistic Regression in identifying fraudulent activities. The results showed that Random Forest and Backpropagation Neural Networks performed exceptionally well in this task, achieving an accuracy rate close to 100%. These findings highlight the significant potential of these advanced machine learning methods to optimize fraud detection systems in financial institutions and provide strong support to further improve fraud detection capabilities.