Research And Application Of Support Vector Machine Multi-classifier

Internet is like a huge data warehouse,which contains text,image,video and other types of data,these data are heterogeneous and unstructured due to different sources.In order to manage and store these data effectively and find useful information quickly,data mining has become a hot research topic as a solution.Support vector machine(SVM)is a commonly used machine learning algorithm in the field of data mining.It has been widely used in the field of classification by virtue of the characteristics of effectively preventing overfitting.In order to solve the common multi-classification problems in daily classification problems,support vector machine produces a variety of different combined multi-classification algorithms.In this paper,two kinds of combined multi-classification structures are deeply studied and analyzed.Two improved support vector machine multi-class classification algorithms are proposed.The main contents are as follows:1.A directed acyclic graph multi-classification(Directed acyclic graph Support vector machines based on fuzzy interval,DAGFSVM algorithm based on fuzzy interval is proposed.In order to solve the problem that noise points often appear in the multi-classification problem of directed acyclic graphs,a fuzzy interval function is set up,which is used to determine whether a sample really belongs to this category,if not,It gives a relatively low fuzzy value and weakens its influence on the final hyperplane.In addition,a separation function is set to measure the accuracy of each class two classifiers,and those with high success rate are placed next to the root node,so as to ensure the accuracy of directed acyclic graph structure as much as possible.2.An adaptive binary tree support vector machine(Binary Tree Support vector machines based on fuzzy interval,BTFSVM)algorithm based on fuzzy interval is proposed.In order to solve the problem of error accumulation in binary tree multi-classification problem,an inter-class distance function is set to measure the relationship between categories,and the similar categories are gathered together,so that the data can construct partial binary tree or approximate complete binary tree according to its own characteristics,and realize the adaptation of the two structures.In addition,fuzzy interval function and auxiliary penalty factor are used to effectively balance the adverse effects of unbalanced distribution in binary tree hierarchical structure classification on the final classification results,which improved the performance of each single binary classification and further improve the accuracy of the overall model.3.The two support vector machine multi-classification algorithms proposed in this paper are compared on different data sets,and their advantages and disadvantages are summarized.Then,based on the advantages and disadvantages of the two algorithms,a simple fund rating system is constructed.The data of the fund rating system is crawled by the crawler from Sina fund data center,and the influence of the value range on the accuracy is eliminated through standardization among the features of the crawling data.Then according to the two different multi-classification algorithms proposed in this paper,we get two different fund evaluation models: fast and accurate,and finally set up two interfaces according to the scale: single evaluation and batch evaluation.Thus,the fund data of different sizes to achieve a simple rating operation.