State Estimation for Hidden Nodes of a Contact Center

A partially observed stochastic network is considered for modelling the operation of a contact center. The network consists of three nodes: the main node being observed is an exponential multichannel finite-capacity queuing system with Poisson arrivals; two hidden nodes are virtual stations for blocked and unsatisfied customers who make repeated calls at a certain intensity. Based on the nonlinear optimal filtering equations, a suboptimal filter is derived for load tracking in the hidden nodes. To provide a comparative analysis of the suboptimal filtering scheme, a drift-based filter is used. In addition, the numerical results of parameter estimation based on the proposed filters are presented.