ATM cash demand forecasting in an Indian bank with chaos and hybrid deep learning networks

This paper proposes to model chaos in the automated teller machine (ATM) cash withdrawal time series of a large Indian commercial bank and forecast withdrawals using deep learning (DL) hybrid DL methods. It also considers influence “day-of-the-week” on results. We first modelled present by reconstructing state space each optimal lag embedding dimension. process converts original univariate into multi variate series. The dummy variable is converted seven variables one-hot encoding augmented multivariate or depending whether was absent. For forecasting future withdrawals, we employed (i) statistical technique namely autoregressive integrated moving average (ARIMA), (ii) techniques such as random forest (RF), support vector regression (SVR), multi-layer perceptron (MLP), group method data handling (GMDH), general neural network (GRNN), (iii) long short term memory (LSTM) network, Gated Recurrent Unit (GRU) 1-dimensional convolutional (1D-CNN). explored 1D-CNN + LSTM GRU. observed improvements forecasts for all when included. that 28 ATMs, whereas remaining 22 ATMs In both cases, yielded best Symmetric Mean Absolute Percentage Error (SMAPE) test data. However, showed statistically different performance than category but equal with non– yet significant 1D-CNN.