Time series forecasting for a call centre

Abstract

Forecasting is an important tool in operations planning. Determining resource requirements and scheduling of personnel at a call centre require accurate forecasts. This study includes analysis and forecasting of the number of calls of general type of customers at a private bank's call centre. These customers are referred to as mass customers and their calls constitute more than 60% of all calls at the call centre. Time series data contain double seasonality with intraday pattern for hourly calls within day and interday pattern within week. The data set of hourly calls of ten-week period is separated into a training set consisting of the first eight weeks and test set consisting of the last two weeks. The methods considered include seasonal Autoregressive Integrated Moving Average (ARIMA) and regression with ARIMA errors. Two models out of different candidate models are identified through statistical inferences of the model parameters, diagnostic checking and fit indices by using the training set for the original series and logged series. Forecasting performances of the models are evaluated with one-step ahead forecasts by using the test set and it is concluded that suggested models adequately predict hourly calls of the call centre. © 2017 IEEE.