The Multi Classification Evaluation Index Of Binary Tree SVM

Support Vector Machine is a supervised learning algorithm of machine learning in the field of artificial intelligence.By mapping,the low-dimensional sample data are mapped to the data in high-dimensional feature space where is a hyperplane separating two kinds of data.Traditional SVM only solves the problem of two-class classification,for multi-classification problem,the multi-classification problem is transformed into multi-binary classification problem,integrating multi-binary support vector machines into multi-class support vector machines.The multi classification SVM of combination methods based on binary tree structure has the advantage of less number of two-class SVMs to be trained,which avoids the occurrence of inseparable and repellent regions of one-to-many and many-to-many methods.However,due to the diversity of the binary tree structure and the feature of guarantee the each sample category is an integral,the proposed construction method based on binary tree structure of multi-class support vector machines lacks specific evaluation criteria for category combination.For the problem of the lack of specific evaluation criteria for category combination in the combinatorial process of multi-classification SVM based on binarytree,in order to ensure that each sample class is a whole,on the basis of the traditional model of information gain ratio based on variable attributes,this paper defines the information gain ratio based on classified attributes,builds the evaluation model of combination structure Based on IG ratio of classified attributes,proposes a evaluation method of combination of BT-SVM based on IG ratio of classified attributes.In the process of constructing binary tree structure,for the left and right two large categories composed of multiple subclasses,this method calculates the IG ratio of classified attributes of each possible combination depending on the variables.For each variable,there exists a maximum IG ratio and the corresponding category combination with the maximum value,the maximum IG ratio of classified attributes of these combinations is used as the criterion to measure the quality of the combinations.Empirical analysis of this method using User Knowledge Modeling and Iris in UCI database shows that the multi class SVM has a high recognition rate as the evaluation criterion value of this combination is maximum.