Abstract:
|
A new methodology is proposed for nonparametric regression when the errors are lognormally distributed. Global confidence bands can also be developed. The methodology then is applied to model the mean service times of calls made to a bank call center. These times are modeled as a nonparametric function of several interesting covariates such as time-of-day, service type, and day-of-week. In service engineering and queueing theory literature, it is quite common to assume that the service times in a service queue will follow an exponential distribution. However, the service times in our data closely follow a lognormal distribution after some precautions are taken. For a queueing system, the mean of the service time distribution is an essential quantity to calculate many basic performance measures. Thus, precise inferences about the mean service time, plus inferences about the arrival process, will enable valid theoretical results, as well as simulations of the real queueing system, in order to help evaluate system staffing and performance.
|