Abstract:
|
Call volume forecasting plays an important role in the operation of call centers. Reliable forecast of call volume enables call centers to allocate resources efficiently and to improve customer satisfaction. In this paper, using a data set detailing the number of calls received by the collection services center of a major telecommunication service provider, the nonparametric transfer function approach is used to explore possible reasons for uneven distribution of daily call volume, to model nonlinear effects that affect daily call volume, and to forecast future call volumes. Polynomial splines are used to model the effects of prior actions taken by the provider on daily call volume for different days of the week and for different risk levels of customers. The serial correlation in call volume is modeled with an ARIMA model. The two-component model is estimated using backfitting. An algorithm is developed to automatically select the models. Forecasting performance is evaluated using post-sample forecasting.
|