Support Vector Machine Based On Rough Set And Multi-granulations

At present, with the rapid development of computer information network,information and data obtained from some fields are increasing, which lead toinformation explosion. Information explosion brought on a data disaster by theincreasing scale of data and information. Now, some technology does not workeffectively any longer, which brought a great challenge to intelligent informationprocessing, so that some new research topics, how to construct a new algorithm to dealwith large data problem, how to use the existing methods to deal with large dataproblem, are proposed.By introducing the neighborhood rough set theory, homomorphism theory, roughset theory and multi-granularity into the support vector machine (SVM), some newalgorithms proposed to deal with massive data are compared with the traditional SVM.The experimental results on some benchmark datasets, not only make the computationalexpense of these algorithms decreased, but also make the classification accuracyimproved.In this paper, the main work is as follows. First, both the rough set theory and themultiple granularity theory are fused into the support vector machine (SVM) theory,and a new classification algorithm, Classification Using Support Vector Machine withRough Set and Multi-granulations (RG-SVM), is put forward. The algorithm is mainlyusing the concept of upper and lower approximation of Rough Set, edge boundariesand the distribution characteristics of support vector to compress the mass dataeffectively and increase the largest interval among samples. The experimental resultsshow that the algorithm can improve the classification accuracy of the sample; Second,both homomorphism theory and multiple granularity theory are fused into the supportvector machine (SVM) theory, thus, an algorithm called Support Vector MachineBased on Multi-granulations (G-SVM) is proposed, which mainly bases on the distribution characteristics of support vector and the multiple granularity theory. Thealgorithm not only makes the computational expense of this algorithms decreased, butalso makes the classification accuracy improved.