Akademik Veri Yönetim Sistemi
10th International Conference on Control, Decision and Information Technologies, CoDIT 2024, Valletta, Malta, 1 - 04 Temmuz 2024, ss.103-108
The forecasting of cash demand for Automated Teller Machines (ATMs) is presented with a formidable challenge, primarily stemming from the dynamic nature of customer cash withdrawal patterns and the resultant fluctuating demand, which is influenced by multiple unknown parameters. An extensive investigation and categorization of the existing body of literature addressing the issue of ATM cash demand are undertaken in this research. Regression, Polynomial Regression, Triple Exponential Smoothing (ETS) and Autoregressive Integrated Moving Average (ARIMA) models are employed as benchmarks. Subsequently, Prophet and machine learning-based models; Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM), are used to predict the daily cash demands. In addition, a de-seasonalizing and de-trending process is adapted, yielding a significant positive impact on the accuracy. The proposed models surpass the findings reported in the literature when applied to the NN5 competition data set consisting of two years of daily cash demand of ATMs across the UK and would have achieved the first place.