A selective classifier for incomplete data

Classifiers based on feature selection (selective classifiers) are a kind of algorithms that can effectively improve the accuracy and efficiency of classification by deleting irrelevant or redundant attributes of a data set. Due to the complexity of processing incomplete data, however, most of them deal with complete data. Yet actual data are often incomplete and have many redundant or irrelevant attributes. So constructing selective classifiers for incomplete data is an important problem. With the analysis of main methods of processing incomplete data for classification, a selective classifier for incomplete data named RBSR (ReliefF algorithm-Based Selective Robust Bayes Classifier), which is based on the Robust Bayes Classifiers (RBC) and ReliefF algorithm, is presented. The proposed algorithm needs no assumptions about data sets that are necessary for previous methods of processing incomplete data in classification. This algorithm can deal with incomplete data sets with many attributes and instances. Experiments were performed on twelve benchmark incomplete data sets. We compared RBSR with the very effective RBC and several other classifiers for incomplete data. The experimental results show that RBSR can not only enormously reduce the number of redundant or irrelevant attributes, but greatly improve the accuracy and stability of classification as well.