Aplicação de Floresta Aleatória para Identificação de Fraudes em Transações Financeiras

Abstract

Fraud detection in digital financial transactions has become a critical challenge with the expansion of e-commerce and online banking. This work explores the application of the Random Forest algorithm to improve the identification of fraud in financial transactions. Using a set of synthetic data generated by the PaySim simulator, it was possible to imitate normal operations and malicious behaviors in financial transactions, overcoming the scarcity of public data. The proposed model focuses on analyzing transaction patterns and early identification of fraudulent activities, employing feature engineering and cross-validation techniques with KFold. The robustness of Random Forest was demonstrated by its ability to deal with unbalanced distributions and mitigate overfitting, achieving an average recall of 99.47%, higher than Logistic Regression and SVM models. Model optimization, carried out via Grid Search, was essential for these results. These findings confirm the effectiveness of Random Forest in detecting financial fraud and highlight the importance of rigorous methodological approaches and advanced machine learning techniques to improve the security and reliability of digital financial transactions.