Aplicação de técnicas de machine learning na identificação de transações fraudulentas no e-commerce

Abstract

With the popularization of the internet over the years, online shopping has become increasingly common, mainly due to the ease and comfort offered, as well as the various payment methods available, with credit cards being one of the main means used in transactions in electronic commerce (e-commerce). However, this convenience also attracts fraudsters who use the internet to perpetrate increasingly sophisticated scams. Thus, it is essential to effectively identify fraudulent transactions to mitigate this risk and protect legitimate buyers. One of the main challenges in fraud detection is data imbalance, as fraud is a rare event and is naturally present in a much smaller quantity than legitimate transactions, which tends to reduce the discriminatory power of the techniques. Among the various Machine Learning techniques commonly used for this purpose, two of them, Logistic Regression and Random Forest, were explored in this work. Additionally, the effectiveness of data balancing was tested using the sampling technique called Undersampling, comparing the two Machine Learning models in scenarios with balanced and unbalanced data. To measure the performance of the models, the Balanced Accuracy metric was used, considered suitable for dealing with imbalanced data. The application was carried out using a dataset obtained from Kaggle, containing artificial data generated through a simulator, with legitimate and fraudulent credit card transactions.