Forecasting cash demand at Automated Teller Machines (ATMs) is one of the most difficult tasks in today's financial system. If an ATM runs out of cash, the bank's reputation will suffer, and the business will incur some expenditure as a result of the decreased usage of these devices by customers. Considering the usage of ‘Any Bank ATM by Any Bank Customer’, the volume of transaction increased multifold and cash availability at all the time is necessary. Banks earn revenue, whenever cross usage of bank’s ATM take place. Looking into the customer service as well as revenue earning perspectives, banks are keen to have right cash availability. It is aimed to have the continued ATM service with neither a customer's transaction is refused due to a lack of cash, nor does idle cash jeopardize the bank's ability to make money from it. The revenue earning is considered one of the yardsticks of the ATM services and thereby A suitable regression machine learning approach to solve the ATM cash prediction model, with the revenue aspect, is taken up as part of emerging trends in computing. This article introduced the Adaptive HindGradiantBoosting Model Regressor (HGBRegressor). The Adaptive HGBRegressor may be utilized to deliver the best and most precise forecast on ATM transactions. That produces high prediction accuracy.

ATM, Adaptive HGB Regressor, Bayesian Optimization, Off-us transactions, hyper tuning, hyper parameters