Research on Multiple Classification Based on Improved SVM Algorithm for Balanced Binary Decision Tree

Support vector machines (SVMs) are designed to solve the binary classification problems at beginning, but in real world, there a lot of multiclassification cases. The methods based on SVM mainly divided into direct and indirect methods, which consist multiple classifiers integrated accordance with certain rules form model, most commonly used present. In this paper, an improved algorithm balanced decision tree is proposed, called IBDT-SVM algorithm. algorithm, it considers not only influence “between-classes distance” “class variance” traditional measures between-classes separability also takes consideration proposes new measure.” Based measure,” finds out two classes largest measure uses them as positive negative samples train learn classifier. After that, according principle class-grouping-by-majority, remaining close these merged classifier again. For uneven distribution or sparse distribution, method can avoid error caused by shortest canter distance overcome “error accumulation” problem existing greatest extent so obtain better According above each layer node traversed until output result single-class label. experimental results show that proposed paper achieve accuracy effectiveness for problems.