An empirical analysis of swarm intelligence techniques on ATM cash withdrawal forecasting

Abstract

Electronic money transfer and its distribution channels such as ATM and on-line/mobile banking have been grown in the last decades. Number of ATMs have had a rising trend since the ATM is cost effective when compared with bank branch. The worldwide number of ATMs increased to 3.6 million in 2017 from 3.5 million in 2015 and estimated to be 4 million by the end of 2021. This re-search is inspired from a real life project which had been undertaken for a large Turkish bank. The project is about determining when to visit and how much to load to each ATM of the bank. The primary question that needs to be answered turns out to be forecasting how much money will be withdrawn from the ATM in the next days. This forecasting task is no way an easy task to cope with since the past withdrawal patterns are too volatile. In essence, the classical time series methods may not perform well on our data. As a result, we decided to implement several swarm intelligence techniques since they are known as well performing optimizers being already practiced successfully to a wide range of problems. Specifically, the techniques utilized in this study are artificial bee colony, differential evolution, migrating birds optimization, particle swarm optimization and simulated annealing. In the study, firstly the aforementioned algorithms are implemented by designing the operators of the algorithms by considering the nature of the problem. Then, parameters of the algorithms are fine-tuned with computational tests. In the last phase, all meta-heuristics are applied to the problem instances with their best performing parameter values and compared through extensive computational experiments.