Bagging Improves Uncertainty Representation in Evidential Pattern Classification
Jérémie François, Yves Grandvalet, Thierry Denceux, Jean‐Michel Roger · Studies in fuzziness and soft computing · 2002
Uncertainty representation is a major issue in pattern recognition when the outputs of a classifier do not lead directly to a final decision, but are used in combination with other systems, or as input to an interactive decision process. In such contexts, it may be advantageous to resort to rich and flexible formalisms for representing and manipulating uncertain information, such as the Dempster-Shafer theory of Evidence. In this paper, it is shown that the quality and reliability of the outputs from an evidence-theoretic classifier may be improved using an adaptation from a resample-and-combine approach introduced by Breiman and known as "bagging". This approach is explained and studied experimentally using simulated data. In particular, results show that bagging improves classification accuracy and limits the influence of outliers and ambiguous training patterns.