Prototype, nearest neighbour and hybrid algorithms for time sreries classification

Holistic classification methods are presented, i.e., similarity measures will be used but no description of the curves by feature vectors of fixed length. Three different methods are adapted: constructionof prototypes for a class kNN classifier and a hybrid of the first two methods, keeping their advantages and overcoming their disadvantages. In a preprocessing step the treated curves are approximated by spline functions. In the case where the measurements are taken from different time intervals, the curves are mapped onto symbol strings.