Detecting fraudulent credit card transactions using artificial intelligence: A methodology

Subject. This article discusses the use of modern machine learning methods to identify suspicious transactions and prevent financial losses in the banking sector. Objectives. The article aims to conduct a comparative analysis of modern machine learning methods used to identify fraudulent transactions with credit cards in order to identify the most effective approaches that can be implemented in the practice of banks to improve the accuracy and automation of the fraud detection process. Methods. The proposed methodology involves comparing various machine learning algorithms to select the most effective one to ensure the accuracy and reliability of anti-fraud systems for the purpose of automated detection of fraudulent credit card transactions. The study examined key machine learning algorithms used in fraud detection tasks, such as neural networks and Random Forest, as well as pre-processing and class balancing methods. The models were trained and tested on real banking data obtained from the Kaggle platform and including 30 predictors and more than 25,000 records. The effectiveness of the algorithms was assessed based on metrics such as AUC-ROC curves. Results. The results of the study show that algorithms based on artificial neural networks and the Random Forest algorithm provide high accuracy in detecting fraudulent transactions. It is found that the combination of these methods with various data pre-processing techniques can significantly increase the effectiveness of anti-fraud systems. In particular, the use of SMOTE significantly improves the accuracy of predictions for the minority class. Relevance. The proposed methodology can be successfully used in the banking sector to increase the reliability of anti-fraud systems. The results obtained may be useful for further improving automated fraud detection systems and minimizing financial losses associated with fraudulent transactions.