Structural Classifier Ensembles for Vector Space Embedded Graphs

The motivation of classifier ensembles is that errors of an individual classifier can often be compensated by the other ensemble members. In the present paper we introduce a general approach to building structural classifier ensembles, i.e. classifiers that make use of graphs as representation formalism and include strings and trees as special cases. The proposed methodology is based on graph embedding in real vector spaces by means of prototype selection. This selection is performed randomized eta times such that the procedure leads to eta different graph embeddings. Hence, a classifier can be trained for each embedding and the results of the individual classifiers can be combined in an appropriate way. In the present paper we take into account that not only the prototypes themselves but also their number has a critical impact on the classification accuracy in the resulting vector space. In several experimental results we make investigations on the classification accuracy of the resulting classifier ensembles and compare them with single classifier systems.