Fraud detection in financial transactions using Machine Learning algorithms

Abstract

The paper addresses financial fraud in credit card transactions. The objective is to evaluate machine learning models that identify the most robust algorithm for the detection of financial fraud. As a solution, a comparison between Random Forest, K-Nearest Neighbors and Decision Tree was implemented by applying the SMOTEEN technique to balance the classes. Two applications were developed, a functional web for data evaluation and visualization, and a desktop for data anonymization. The important results showed that Random Forest's model was the most balanced, obtaining outstanding metrics such as 99% Accuracy, 99% Precision, 98% Recall and 99% F1-Score with an ROC AUC of 0.99%. The proposed approach, based on ensemble models and class balancing techniques such as SMOTEEN, proves to be an effective and adaptable alternative to strengthen financial fraud monitoring systems.