Development of a generalized algorithm for identifying atypical bank transactions using machine learning methods

• Published in: Procedia computer science Vol. 213; pp. 101 - 109
• Main Authors: Domashova, Jenny, Kripak, Elena
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2022
• Subjects: anomaly search algorithms; atypical transactions; class imbalance elimination methods; classification; credit institutions; gradient boosting; machine learning methods; anomaly search algorithms; credit institutions; class imbalance elimination methods; gradient boosting; machine learning methods; classification; atypical transactions
• ISSN: 1877-0509; 1877-0509
• DOI: 10.1016/j.procs.2022.11.044

The article presents the results of the development of an algorithm for identifying bank transactions that are not typical for a client, based on data pre-processing and machine learning methods. Various methods of classification have been studied while modelling this problem. The procedure for forming a training sample in the absence of labelled data was carried out. A generalized algorithm for detecting atypical banking transactions has been developed, which is based on the search algorithms for anomalous objects such as Isolation Forest, Elliptic Envelope and Local Outlier Factor, the neural network architecture Autoencoder and the XGBoost classifier, trained on a sample with the eliminated imbalance using the Tomek Links Method. The developed algorithm can be used by banks to timely identify transactions that are atypical for a client based on data analysis, which will assist in minimizing reputational and financial losses of banks, as well as strengthening the financial and economic security of their users.