ATM Cash Flow Prediction Using Local and Global Model Approaches in Cash Management Optimization

Cash management optimization is one of the most essential tasks for any bank, because it helps save a significant amount of money by reducing the cost of ATMs funding and encashment. This paper focuses on forecasting customer cash demand, which is one of the key components of the optimization system. Furthermore, for the first time, our research touches on the problem of nonstationarity, which is typical for real-world ATM data, and proposes a data preprocessing pipeline to tackle it. We proposed new forecasting methods in the paradigms of local and global models, proved their superiority over classical approaches to forecasting time series and approaches used specifically for the cash demand forecasting problem.