Detecting Fraudulent Transactions Using a Machine Learning Algorithm

This study is dedicated to the problem of rapid detection of fraudulent financial transactions. Current approaches to monitoring and detection of fraud in banking transactions were analyzed. The problem of the most reliable recognition of classes of financial transactions using unbalanced data was considered. The problem of choosing the best classifier among ensemble algorithms was investigated. Specifically, these are algorithms of a Random Forest, Adaptive Boosting, and Decision Trees bagging. Methods of solving the problem of unbalanced data samples were analyzed. It was suggested to use the random undersampling algorithm to create balanced subsets of ensemble algorithm estimators. The results of experimental comparison of the selected methods are presented.