Neural Networks Supporting Causal Reasoning in Traffic Telematics

W. Toplak, J. Asamer, and K. Din (Austria)


Data mining, neural networks, genetic algorithms, causal reasoning, intelligent information systems


In recent years neural networks (NN) made their way to the science of traffic telematics. Approaches for their implementation are quite different, depending on the amount of available data sources and on data resolution (i.e. sampling times). Recently the trend goes to fuse different aspects of system observations to make decisions more secure and reliable. A reason for that is the increasing availability of various data coming from different sources (e.g. weather, events). Weather, concerts as well as higher traffic flows caused by vacationists occur timely and spatially separated. The dimension of designable models grows and when a traffic indicator such as flow needs to be mapped adequately there is additional and meta-information which can be included in the process of system perception. Two approaches are introduced where NN are used to map a given system with the goal to classify or predict a traffic flow, respectively. Observation data of this indicator was only available for training and testing. Decisions are based on additional and meta-information, named as the ability for causal reasoning. The first approach uses feed-forward NN, the second approach uses a Self-Organizing Map (SOM), leading to better visual causal understanding.

Important Links:

Go Back