Merkmalsbildung und lernfähige Klassifikation von Zeitreihen

In the paper classification methods for time series with a high number of measured values are developed. The measurements can be noised. Therefore, the use of classification methods is not possible without former treatment. In order to form and reduce features, the time series are approximated by spline functions where the number and the location of the joints are defined by the method. If the measurements are given over one and the same time interval, the splines can be classified immediately. Otherwise, the splines can be transformed into strings. For classifying strings, a distance comparing strings of different lengths is defined with the help of the so-called compatibility graph.