支持向量机多分类方法研究及其在基金评价中的应用

发布时间：2018-02-13 14:09

本文关键词： 多分类支持向量机基金评价　出处：《北京交通大学》2014年硕士论文　论文类型：学位论文

【摘要】：支持向量机是基于统计学习理论解决机器学习问题的一种新工具,它最初由Vapnik及其同事于20世纪90年代提出,随着统计学习理论的完善,近年来其在算法研究及实际应用方面都取得了突破性进展,其中多分类支持向量机的研究更是当前的热点。本文所做的主要工作有：一、对支持向量机多分类算法的理论基础与算法思想进行了总体阐述,并对各种多分类算法的性能进行对比研究。针对现有多分类算法存在“拒分区域”、SVM“错误累积效应”的不足,提出了改进后的基于二叉树的多分类支持向量机。本文创新性的引入了类分散度的概念,构造最佳分类顺序二叉树结构,改进现有算法在推广能力方面的缺陷,经过算法实现并带入UCI数据集验证,新算法在分类精度和时间上表现出优于传统的OAA-SVMs、OAO-SVMs、DT-SVMs算法。二、详细总结了国内外现有基金评级方法,介绍比较经典的基金评价方法体系,并对现有的评价体系的优缺点进行总结评述。三、建立基于多分类支持向量机的基金评价模型。以基金各项评价指标作为输入向量,使用主成分分析法进行特征提取,并选择适当的核函数及通过交叉验证获得最佳模型参数,对训练集进行训练得到分类函数,再使用预测样本集进行验证分析得到预测样本基金的评级。此外,本文使用两种多分类支持向量机算法——OAA-SVMs与改进DT-SVMs进行基金评价实证分析,分类准确度最高达到80%。结果表明支持向量机方法对我国基金业绩评价有良好的可行性和有效性,并且改进DT-SVMs在分类精度和分类时间上都优于OAA-SVMs。
[Abstract]:Support vector machine (SVM) is a new tool for solving machine learning problems based on statistical learning theory. It was first proposed by Vapnik and his colleagues in 1990s, with the improvement of statistical learning theory. In recent years, it has made a breakthrough in algorithm research and practical application, among which the research of multi-classification support vector machine is a hot spot. The main work of this paper is as follows:. First, the theory foundation and algorithm idea of support vector machine multi-classification algorithm are introduced. A comparative study of the performance of various multi-classification algorithms is carried out. In view of the shortcomings of the existing multi-classification algorithms, such as "rejection region" and "error accumulation effect" of SVM, This paper proposes an improved multi-classification support vector machine based on binary tree. This paper innovatively introduces the concept of class dispersion, constructs the structure of the best classification sequence binary tree, and improves the shortcomings of the existing algorithms in popularizing ability. The new algorithm is better than the traditional OAA-SVMsOO-SVMsN DT-SVMs algorithm in classification accuracy and time. Secondly, the paper summarizes the existing fund rating methods at home and abroad, introduces the classical fund evaluation method system, and summarizes the advantages and disadvantages of the existing evaluation system. Thirdly, a fund evaluation model based on multi-classification support vector machine (SVM) is established. The evaluation indexes of the fund are used as input vectors, and the principal component analysis (PCA) is used for feature extraction. The best model parameters are obtained by selecting appropriate kernel function and cross validation. The classification function is obtained by training the training set, and the rating of the forecast sample fund is obtained by using the forecast sample set to verify and analyze. In this paper, two kinds of multi-classification support vector machines (OAA-SVMs) and improved DT-SVMs are used to carry out the empirical analysis of fund evaluation. The results show that the classification accuracy is as high as 80%. The results show that the SVM method is feasible and effective for fund performance evaluation in China. And the improved DT-SVMs is superior to OAA-SVMsin classification accuracy and classification time.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP18;F832.51

【参考文献】