An adaptive fuzzy neural network for traffic prediction

This paper proposes the use of a self-adaptive fuzzy neural network for traffic prediction. The necessity of using a self-adaptive predictor arises from the time shifting nature of probability distributions in an urban traffic network. We advance the use of an architecture which tracks these changes over time, taking into account distribution drifts due to weather conditions, season, or other factors. Tests are run over a synthetic data set which emulates the change in dynamics for an arc in a traffic graph. We introduce the use of a pruning procedure with re-training over the test and cross-validation sets, followed by prediction over short time horizons.