ATM cash stock prediction using different machine learning approaches

One of the most common problems related to banking systems is the Automated Teller Machine (ATM) cash demand forecasting. Cash shortage adversely affects customer satisfaction, while too much cash reduces bank’s profitability. We have developed an ATM cash prediction system using different traditional statistical and machine learning approaches, including linear regression, support vector machines, artificial neural networks, LSTMs and traditional statistical analysis (ARIMA) on the same ATM data. We compared the results of these methods and showed that machine learning methods in comparison with ARIMA have higher accuracy. Also it was shown that among the machine learning models, LSTM gives the most accurate predictions and use less features compared to other models