Profitability Prediction for ATM Transactions Using Artificial Neural Networks: A Data-Driven Analysis

Abstract

Banks are tended to increase their transactions income against the costs of an ATM, which includes installation, setting up and maintenance. Due to dependence of income on various factors such as geography and demography, ATM’s income function is complex. Therefore, understanding its behavior could lead to discover the mathematical relation between ATM installations weighted variables versus ATM’s profitability. In this study based on artificial neural networks (ANNs) prediction model and the real data of 374 ATMs in Tehran, a comprehensive income model is presented which can predict the profitability of an ATM. In order to have accurate analysis and better training for ANNs, statistical methods are used to find out the correlation between ATM installation variables and its profitability. Results show that the feed-forward and Elman networks can predict the income of transactions with minimum error. Applying these analyses will help banks in making optimized decisions to provide ATM services to customers.