On the prediction of dispenser status in ATM using gradient boosting method

Abstract

The purpose of this study is to solve the problem of determining the status of an ATM to increase its resiliency, which will reduce the reputational and financial losses of banking structures. The tools for the goal achieving were machine learning methods such as gradient boosting model (on the example of Russian Sberbank ATM data). The study showed good enough accuracy in the problem of binary classification of the status of ATM dispensers, and also revealed the importance of time characteristics of the occurrence of errors and transactions. The methodology has practical value; it is possible to use it outside the banking sector due to the flexibility of the chosen model.The purpose of this study is to solve the problem of determining the status of an ATM to increase its resiliency, which will reduce the reputational and financial losses of banking structures. The tools for the goal achieving were machine learning methods such as gradient boosting model (on the example of Russian Sberbank ATM data). The study showed good enough accuracy in the problem of binary classification of the status of ATM dispensers, and also revealed the importance of time characteristics of the occurrence of errors and transactions. The methodology has practical value; it is possible to use it outside the banking sector due to the flexibility of the chosen model.