Rough Sets Theory And SVMs Based Multi-class Classification Algorithm

As a currency learning techniques introduced recent years,SVMs which handle small sample size problem have the features of good generalization ability,solid theoretical background,high accuracy ,and getting global optimization.Ho\vever,it is a classifier for two-class originally and not suite for the multi-class problems and dealing with large data sets .On the other hand,,RST has the features of handling and reducing large data sets while has lower classification accuracy than SVMs.In this paper ,the data are classified in advance with the RST,and two methods of combination of the data to classify two-class problems are proposed. Then one against one classification is performed with SVMs.So the muti-class problem can be solved,the accuracy of classification guaranteed ,and the reduction of the data carried out .In particular ,the approach to classification based on the equivalence classes of the main attribute is explicit conceptually, easy to understand and implement.Furthermore,the reduction of the sample size is distinct.The approach to classification based on the equivalence classes of the main attribute is as follows:(1) Discretize the continuous training data.(2) Remove the indiscernible attribute.(3) Conduct attribute reduction and value reduction .(4) Choose the attribute in which the number of the equivalence is largest as the main attribute.(5) Take every equivalence as a subset and train it with SVMs.(6) Test using decision function.Further more ,the experimental results are satisfying.