Traffic information imputation using a linear model in vehicular ad hoc networks

Vehicular ad hoc mobile networks are very promising platforms from which data for traffic prediction algorithms can be extracted. However, they present unique challenges, because the quantity, quality, and original location of the data are essentially random variables. This is much different than traditional platforms where fixed sensors at known locations provide known quantities of data at fixed intervals. In this paper, a linear model is proposed to impute missing information, based on the spatial relevance of candidate traffic information. The example application is the prediction of space mean speeds on a target link in a rectangular highway network, although other applications can be supported with the same ideas. The space mean speed was estimated through 15 consecutive one-minute space mean speed sets. As the result of computer simulation experiments, one of four target links was excellently represented for all times and two others at a reasonable level. A fourth link was not very dependent on surrounding links. In conclusion, because traffic conditions on a target link are associated with those on neighboring links, and seem to be affected by a compound relation with those links, it is expected that it is possible to estimate missing information through the linear model with a composition of information on neighboring links