In this paper we present a robust mode selector for the uncertain graph-constrained Switching Mode Model (SMM), which we use to describe the highway traffic density evolution. Assuming an uncertain speed of the congestion wave, the proposed selector relies on a transition digraph suitably incorporating the present and historical statistical traffic information, to determine the most probable current mode of the SMM. Its effectiveness is demonstrated on the problem of traffic density reconstruction via a switching observer, in an instrumented 2.2 km highway section of Grenoble south ring in France.