Short-term forecasting of travel time is one of the central topics in current ITS research and practice. The most widely applied travel time forecasting approach is the neural network. Usually many candidate neural networks are trained and the network performing best on an independent validation dataset is selected. However, the training data then needs to be divided in two, leading to less well trained networks. Using Bayesian inference theory, a selection criterion called the `evidence' can be derived for each network without the need for a validation set. This results in higher prediction accuracy as more data can be used for training, Moreover, a committee of neural networks can be constructed using the evidence. A case of forecasting travel times on the A12 motorway in the Netherlands shows that the committee approach indeed leads to improved travel time forecasting accuracy, and that the evidence should be preferred over the validation set approach when constructing the committee.
Neural Network Committee to Predict Travel Times: Comparison of Bayesian Evidence Approach to the Use of a Validation Set
2008-10-01
231758 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Bayesian committee of neural networks to predict travel times with confidence intervals
Online Contents | 2009
|Bayesian committee of neural networks to predict travel times with confidence intervals
Online Contents | 2009
|