Implementation Of SVM Optimized By Multi-objective Optimization Based On Pareto

Classification Learning is the important content in Machine Learning. SVM (Support Vector Machine) has a good performance of classification compared with other classification algorithms. SVM can have different performances of classification through choosing different Kernel Functions and parameters, and this characteristic makes it flexible in applications. At the same time, there is a necessary relationship between SVM's performance and Kernel Function and parameters. SVM can not solve multi-classification efficiently like other binary classification algorithms. The system uses AUC (Area Under the ROC Curve) evaluation standard which can make up the disadvantages of correct rate. Multi-objective optimization improves the unilateralism of single objective evolution in considering multi-objective.It implements multi-class classification to combine AUC evaluation standard and SVM multi-class classification in this thesis. It optimizes 1/E(AUC) and D(AUC) through multi-objective optimization based on Pareto, and designs and implements a new algorithm named SVM_PARETO (SVM Optimized by Multi-Object Optimization Based on Pareto).