A New Construction Method of Heuristic Multi-Class Hierarchical Tree Structure Model Based on Support Vector Machine and Information Entropy

SVMs have limitations in the process of solving multi-class classification problems, often accompanied by many problems such as data set skew, excessive base classifiers, and poor adaptability of coding methods, resulting in a significant drop in accuracy and efficiency. This paper proposes a multiclass hierarchical tree structure support vector machine recognition algorithm based SVM and Information Entropy, which can be used in any SVM-based algorithm to reduce training time and improve accuracy. This research selects the BreastTissue data set in the UCI database as the research object, the overall accuracy of the experiment is increased by about 5%, and the training time is reduced by about 29%. This research method provides a new idea for the research of support vector machine multi-classification and it also has certain theoretical reference significance for the research of classification and regression problems in ontology domain.

I. Introduction

Support Vector Machines (SVM) [1] is a machine learning method that has many advantages over traditional machine learning methods, such as good generalization performance, strong sample learning ability, efficient model generalization performance and insensitive dimension. The SVM method itself is designed for the two-category problem. To extend it to solve the multi-class classification problem, it is necessary to transform the multi-class classification problem into several types of problems. The methods commonly used at present are: one-versus-one (OVO) [2], one-versus-all (OVA) [3], error-correcting output codes (ECOC) [4], Minimum Output Coding (MOC), etc. In the process of extending SVM from binary-class classification to multi-class classification, there are inseparable problems in the edge of high-dimensional feature space classification and data set skew. These problems have attracted extensive attention from relevant scholars in recent years.