Resampling based Modelling of Individual Routing Preferences in a Distributed Traffic Network
Jeļena Fiošina, Maksims Fiošins · 2014
We consider a traffic routing algorithm, which requires modelling of individual routing preferences and is based on a selection of the shortest itinerary. This selection leads to route comparison on the basis of historical data and dynamic observations. We propose a generic cloud-based system architecture, based on the collaboration of individual and cloud agents and resampling-based pairwise route comparison in a stochastic graph. The weights of the edges are considered to be independent random variables with unknown distributions. Only historical samples of the weights are available, and some edges may have common samples. We estimate the probability that the weight of the first route is greater than that of the second one. The analytical expressions for the expectations and variances allow theoretical evaluation of the method. To choose from the available alternative routes we apply a four-step decision-making process, instantiated for route recommendations and Markov chain based route ranking method for selection of the final decision. The experimental results demonstrate that the resampling estimates are more precise than parametric plug-in ones in the case of extreme small or extreme large sample sizes.