Estimating traffic data in traffic networks by singular value decomposition and maximum-likelihood

Traffic data is often estimated on basis of historical time series for single location. We propose an adaptive estimator for a whole traffic network on the basis of multivariate statistics. For data analysis Singular Value Decomposition is used. A Maximum-Likelihood-Estimator and current data from selected detectors enable to estimate data from all other detector sites. These methods are applied to data from inductive loops from the City of Nuremburg. Estimation errors are smaller than those of conventional estimation-systems and can be furthermore improved by sensor and data allocation.