2025, Vol. 8, Issue 2

With credit card fraud becoming a major concern, advanced fraud detection systems are required to protect financial transactions due to the exponential growth in credit card transactions. It appears that manually identifying the questionable transaction is very challenging and time-consuming. These issues are resolved by machine learning because of its accuracy and speed. This study has demonstrated the accuracy with which machine learning algorithms identify fraudulent transactions.

Robust models were constructed using algorithms such as Logistic Regression, Decision Tree, Random Forest, LightGBM, XGBoost, Adaboost, and CatBoost using a standardized dataset that contains both authentic and fraudulent transactions. Comprehensive analyses were conducted using a variety of classification criteria, including F1 score, recall, accuracy, and precision. The effectiveness, drawbacks, and advantages and downsides of various algorithms were examined. All of the developed models have been proven to perform better; however, in comparison, models constructed using the Random Forest, XGBoost, Decision Tree, and LightGBM algorithms are more accurate, while CatBoost has produced the lowest accuracy.