Effectiveness of Machine Learning in Fraud Detection. Systematic Literature Review

Resumen

Fraudulent transactions represent a significant global issue due to their economic and social impact. This research aims to identify the most effective machine learning models for detecting fraudulent transactions. A systematic literature review was conducted as the primary methodology, structured around three specific questions derived from the main research question: Which machine learning models are the most effective for detecting fraudulent transactions? A total of 78 articles were analyzed, extracted from the Scopus and Web of Science databases up to September 2024. Of these, 39 met the inclusion criteria established under the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) protocol. The results highlight that machine learning and deep learning models, such as Random Forest (RF), Extreme Gradient Boosting (XGB), Long Short-Term Memory (LSTM), Artificial Neural Networks (ANN), and some hybrid ML-DL models, are the most effective for detecting fraudulent transactions. It is concluded that these techniques provide robust and reliable solutions to prevent losses caused by fraud, contributing to the development of advanced strategies in organizations and financial sectors.