Short-term forecasting based on a transformation and classification of traffic volume time series

This article describes a short-term forecasting method for traffic volumes at cross-sections. The prediction is based on classified historical patterns. Continuously collected time series are transformed into a representation with objects. These objects are interpreted as polylines and support the qualitative and quantitative interpretation of the data. The historical patterns are classified in great detail, including environmental conditions and the shape of the patterns. An evaluation based on collected data from a field trial in Köln is presented. The results of the new approach are compared with those from two simple time series predictors. The sensitivity analysis shows the differences of using different levels of classification.