Optimization Of Multi-class SVM Algorithm Based On AUC

SVM algorithm has been widely used in classification learning domains because of its high accuracy. There is no a uniform model for the classification method of SVM in multi-class and cost-sensitive domain, and accuracy cannot ensure the total misclassification cost least. AUC (Area Under the ROC Curve) has been widely used to evaluate the multi-class classification algorithms in machine learning as it measures the total performance of classification algorithm when the dataset is of unbalanced classes or the cost of misclassification is unknown. This thesis combines the AUC and multi-class SVM method to classify the multi-class data, and then with GA (Genetic Algorithm) to optimize the classification result AUC. The proposed algorithm GOSMAUC (GA Optimizes Multi-class SVM Based on AUC) is used to evaluate classification result of the multi-class SVM and optimize the AUC.GOSMAUC algorithm is developed based on LIBSVM platform. Empirical results indicate that GOSMAUC effectively evaluates the classification performance of multi-class SVM, and the AUC appears effective to the previous algorithms. This algorithm has good calculating performance, better comprehension and deals with skew dataset. This algorithm can be used to solve the multi-class classification problems.