Comparison of different information fusion methods using ensemble selection considering Benchmark data

The overall system reliability of complex or safety critical systems is of increasing importance in a lot of application fields. To ensure a high accuracy of the decisions evaluating situations or conditions, the assignments from different classifiers can be fused to one final decision. The kind of selection as well as the individual properties of suitable classifiers to be fused are crucial for the overall accuracy. In this contribution a detailed comparison of three different fusion methods (Weighted Voting, Bayesian Combination Rule, and Dempster-Shafer Combination) using different ensemble selection strategies (Static Classifier Ensemble and Dynamic Classifier Ensemble) is given. Using a set of Benchmark data the results are numerically analyzed concerning the number and quality of classifiers to be combined. Based on this (general) example questions about a suitable combination of ensemble selection and fusion method can be answered.