Data fusion using feature selection based causal network algorithm

We propose a statistical definition of reduct and develop a feature selection algorithm based upon it. It shows that the features found by this algorithm get the largest coverage of the objects, and is most resistant to noise compared with the results found by genetic and dynamic reduct searching algorithm when they are applied to a water-pollution monitoring multisensor fusion system, which is described by the causal network model. Comparative tests show that with the selected features, the efficiency of the causal network based searching algorithm is greatly improved, at the same time the classification accuracy is maintained.