Detection of Fraud in Credit Card Transactions

Detecting fraud in credit card transactions is a major strategic challenge for financial institutions, given the rapid growth of electronic payments and the steady increase in the volume of data generated. Traditional control approaches quickly reach their limits in the face of the growing complexity of fraudulent behavior and the significant imbalance between legitimate and fraudulent transactions. This study analyzes the combined contribution of big data and machine learning techniques to improving the automatic detection of bank fraud. The study is based on Kaggle’s public “Credit Card Fraud Detection” dataset, which consists of 284,807 transactions, only 492 of which are fraudulent. Several machine learning algorithms were implemented and compared, including logistic regression, Naive Bayes, K-Nearest Neighbors, decision trees, Support Vector Classifier (SVC), Random Forest, and XGBoost. Particular attention is paid to data preprocessing, handling class imbalance, and selecting appropriate evaluation metrics, such as accuracy, precision, recall, F1-score, ROC AUC, and PR AUC. Experimental results show that linear models exhibit limited performance, particularly due to a high number of false positives. In contrast, ensemble models demonstrate superior discrimination capabilities. The XGBoost model achieves the best performance with an AUC-ROC of 0.97 and an F1-score of 0.871. Random Forest, with an AUC-ROC of 0.94, has a precision of 0.934. SVC also demonstrates good class separation capability with an AUC-ROC of 0.97, but at the cost of higher computational cost and a precision of 0.248. The integration of Spark and Kafka technologies, combined with high-performance supervised learning algorithms, helps improve the reliability and integrity of payment systems, while paving the way for more robust, scalable fraud detection solutions that are tailored to real-world operational constraints.