System analysis, management and information processing

Building a machine learning model for predicting fraudulent transactions

A. F. Konstantinov, L. P. Dyakonova

Plekhanov Russian University of Economics, 115054, Russia, Moscow, 36 Stremyannyy lane

Abstract: The article presents development of a machine learning model for predicting fraudulent transactions using transactional data from a bank. It discusses the features of encoding categorical variables related to the presence of time in the transactional data to avoid information leakage. Additionally, experiments were conducted on the application of bagging and the creation of additional variables based on their contribution to the final prediction using Shapley values. The quality metrics of the machine learning model are examined and analyzed.

Keywords: fraudulent transactions, catboost, encoding categorical variables, catboost_encoder, target_encoder, bagging, variables creation, Shapley values

UDC: 004.89

MSC: 90С99

Received: 16.01.2025
Revised: 27.01.2025
Accepted: 10.03.2025

DOI: 10.35330/1991-6639-2025-27-2-11-22

Bibliographic databases:

•