Modelling and forecasting of call center arrival process

Fitting arrival process and forecasting future arrivals are crucial to staffing and scheduling in call centers. According to different stages of a call center, arrivals are classified into IVR arrival and agent arrival. Non-homogeneous Poisson process has been widely used overseas for modeling stochastic agent arrival process. However, this study initially proposed IVR arrival fitting and forecasting. The IVR arrival process of this call center appears to be “overdispersed” when comparing the mean arrival rate and its variance with the corresponding Poisson process. Therefore, time series was used to model and predict the IVR arrival process. White noise test of residuals was applied and the MAE (mean absolute error) was adopted to evaluate the goodness of fit. The results show that ARIMA (1,0,1) is preferable for predicting the IVR arrival in a short period of normal days and Winters is preferable for the Spring Festival period. Finally, the regression method was employed to describe the relationship between IVR arrival and agent arrival, and predict the agent arrival.