Comparative analysis of machine learning models for the detection of fraudulent banking transactions

This research presents a comparative analysis of machine learning models for detecting fraudulent banking transactions, a growing problem in the digital financial sector. The aim is to evaluate and determine the most effective model for identifying suspicious transactions, overcoming the challenge of a highly imbalanced dataset. Using data from 565,000 real-world transfers, models based on algorithms such as Random Forest, Neural Networks and Naive Bayes were built and tested. Of these, the Random Forest model proved to be the most robust, achieving 100% accuracy for legitimate transactions and 95.79% accuracy for fraud detection. This level of accuracy reinforces its viability for implementation in real-time banking systems. This result underlines the feasibility of integrating such a model into banking systems for real-time analysis, allowing the interception of dubious transactions. This capability would drastically reduce exposure to financial fraud, optimizing transaction security without compromising operational fluidity. In addition, limitations were identified and future approaches discussed, including oversampling techniques and adjusting class weights to improve detection in unbalanced datasets.