Waiting-Time Estimation in Bank Customer Queues using RPROP Neural Networks

Abstract

In daily banking customer queues, unknown waiting-time could lower customer experience. Little’s Law formula in Queue Theory provides a generic formula for waiting-time, but it cannot be implemented directly to give finite wait-time estimation in real-life. This study aims to investigate predictive variables that explain waiting-time duration. This paper uses Fast Artificial Neural Network engine to implement Artificial Neural Networks method. To train Artificial Neural Networks, Resilient Propagation was used. Time-series approach and structural approach for input neuron was compared. Average duration from previous interval and number of server was proposed to increase structural variable like Queue Length and Head of Line Duration estimator variable. To determine the best configuration for number of neuron in input and hidden layer, experimental method was used. The results of this study show that structural approach provides better estimation than time-series approach. Furthermore, modified helper variable combination provides a more refined result.