Estudo comparativo de modelos de machine learning aplicados ao contexto de detecção de fraudes em cartões de créditos

Abstract

According to data made available by the Central Bank of Brazil over the years, the use of credit cards in the country has been becoming more popular, making it a target for a significant number of frauds. Credit card fraud can be defined as the unauthorized use of the payment method, which can occur with or without the physical card being used, given the possibility of data theft through cyber attacks. In this context, the main objective of this work is not only to present the problem but also to conduct a comparative study of different machine learning techniques to automate the process of fraud detection in credit card transactions. Among the analyzed techniques are Random Forest, Logistic Regression, Decision Tree, Naive Bayes, and Multi-Layer Perceptron (MLP). Regarding the performance analysis of the models, the following indicators were used: accuracy, Mathews correlation coefficient (MCC), F1Score , the area under the Precision-Recall curve (AUC-PR), training time, and the time used by the models to classify new samples. As a result, it was found that the Multi-Layer Perceptron (MLP) model had the best performance, both in terms of detecting fraudulent and genuine transactions and in terms of the time spent during the training and classification processes of new samples. This metric is crucial due to the existence of scenarios where thousands of transactions are made simultaneously. In conclusion, the feasibility and relevance of using different machine learning and data processing techniques in combating credit card fraud are exciting points to analyze, indicating promising paths for future advancements in financial security.