Sample Selection Algorithms Based On Sample Entropy And Pre-clustering

Feature selection is one of important step in pattern classification. By feature selection,the redundant features can be removed, the computational complexity can be decreased andthe classification accuracy can be increased. Similarly, there exist redundant and unimportantsamples (also named instances) in the datasets which have no contribution to classification.Recently, sample selection is hot topic in machine learning and draws many researchers’attention. By sample selection, we can eliminate the redundant information (instance) in thedatasets; select more important and fewer samples as training set to train classifier, whichmay have no loss information contained in the original datasets. Accordingly, theclassification accuracy and generalization ability can be improved while the computationalcomplexity can be decreased.By studying the condensed neighbor rules of Fuzzy KNN, we propose an sampleselection algorithm based on sample entropy in uncertainty circumstance in this paper. Inaddition, we propose another two stages sample selection algorithm based on pre-clustering.Some experiments are conducted on artificial dataset and UCI datasets. The experimentalresults demonstrate that the proposed methods are feasible and effective.