Improved SVM for Multi-class Classification by fuzzy game theory

SVM is one of the popular classification algorithms based on statistics learning, which is presented for two-class problems. In real environments, the problem is usually multi-class. Thus, multi-class separation methods are very important compared to binary classes. In this work, to decrease the complexity of the model and the resulting loss of accuracy, fuzzy game theory is derived, which will be able to map the non-linear to a linear problem. Fuzzy game theory is obtained from the probability of data in each class by using two players (in our problem, each player is equivalent to a class label). Here, the decision matrix is yielded by the fuzzy logic, and then the equations are solved by the linear programming. Obtained results from the computer simulation validate the SVM model by fuzzy game theory.