Multiple sources data fusion strategies based on multi-class support vector machine

Abstract

Data fusion from multiple sources is very important and useful. Some strategies for multiple sources data fusion based on Multi-class Support Vector Machine (MSVM) are proposed in this paper. The features (independent components) of data from multiple sources are extracted for fusion. The Dempster-Shafer theory (DS theory) and Bayesian theory are used to combine the probabilistic outputs of MSVMs. Then the outputs of DS theory combination are classified by a MSVM to get the final decision of the classification. Finally, these strategies are evaluated by three data sets and the results show that DS theory can improve the accuracy obviously, and the strategies based on MSVM and DS theory is very fit for solving problems with small data sets. © 2008 Springer-Verlag Berlin Heidelberg.