Fraud analysis prediction in online credit card transactions using machine learning methods

Credit cards have rapidly integrated into people's daily lives and become one of the securely used payment methods worldwide, thanks to their widespread usage and robust infrastructure. However, the increasing number of credit cards and the rapid growth in transaction volume have attracted fraudsters, leading to the emergence of various fraudulent methods aimed at gaining unjust profits. The ease of accessing credit card information today facilitates the activities of credit card fraudsters. With advancing technology, account transactions can be analyzed over time, enabling the tracking of the use of malicious data. In this study, utilizing the Credit Card Fraud Detection dataset obtained from Kaggle, significant findings are presented by comparing a community-based XGBoost model with other traditional machine learning models for detecting credit card fraud. It is noted that XGBoost, Random Forest, and CatBoost models perform better on performance metrics such as precision and recall. These results indicate the potential of reducing fraud risks encountered in the daily operations of financial institutions. A noteworthy point of the study is the emphasis on the ability of XGBoost, Random Forest, and CatBoost to handle imbalanced datasets. In cases where traditional models exhibit poor performance, XGBoost, RandomForest, and CatBoost are reported to achieve up to 99% prediction accuracy and provide better performance compared to other models. Keywords: Multilayer Neural Networks, Data Mining, Naive Bayes Classifier, Fraud Detection.