Parameter estimation accuracy and active learning in the zero-range process

Analysis of traffic flow is one of the prominent concerns in traffic engineering, and it seeks to both elucidate the generation process of traffic jams and to ease them. A zero-range process (ZRP) is a representative traffic-flow model described as a probabilistic cellular automaton. There is a parameter that controls the behavior of the vehicles in the model, and parameter estimation enables us to determine the unobservable behavior from the observable flow data. There are few studies on estimating the parameter, but the properties of models with a known parameter have been well investigated mathematically. In the present paper, we determine the accuracy of parameter estimation and propose an optimization method to collect the data, which corresponds to active learning.