Skip to Main Content
In classification problems where additional information is obtained from different data sources having heterogeneous features, a classifier trained earlier on the original data can not be updated to learn new data. One solution is to fuse this data by using classifiers to learn from each source and to combine outputs in order to get a more accurate final decision. Ensemble methods are well suited to solve data fusion problems since they use all data sources available while taking advantage of complementary information. Robust methods are proposed for combining classifiers, aimed at reducing the effect of outlier classifiers in the ensemble. The proposed methods are shown to have better performance leading to significantly better classification results than the previously employed techniques.