半监督SVM在阿尔茨海默症数据分析中的应用

发布时间：2018-08-27 18:31

【摘要】：阿尔茨海默症(AD)是一种以认知障碍为主要临床特征的慢性病,也是老年疾病中的常见高发病。随着生物医学的发展,对阿尔茨海默症的研究数据越来越多,但是这些数据集具有高维、形式多样以及分布不平坦的特点,如何有效地利用这些复杂的数据成为当今大数据时代所要研究的热点问题。支持向量机(SVM)是基于统计学习理论发展起来的,是借助最优化方法进行数据挖掘的一种新工具。但该方法不能够识别模糊标记样本,也不能利用未标记样本,导致模型分类结果出现偏差。为了有效地处理阿尔茨海默症中的复杂数据,且不浪费大量有价值的未标记样本,引入了支持向量机的改进算法,即模糊支持向量机(FSVM)和半监督支持向量机(S3VM),并将这两种方法应用到阿尔茨海默症数据的分类技术中,通过实验观察分类结果的准确性。主要研究内容和结果如下:(1)首先,采用了特征提取方法对数据进行前期处理。为了降低数据的维度,利用主成分分析法从121个阿尔茨海默症样本数据的55项特征变量中提取出11个因子变量,且这些因子变量基本上能够代表数据的全部信息;(2)研究了支持向量机的理论框架,针对支持向量机模型中的核函数与参数问题,通过设定不同的值进行分类实验,观察其对分类准确率的影响程度。实验结果表明,SVM算法能够对阿尔茨海默症数据进行有效的分析,并且测试样本的分类准确率可达92.157%;(3)研究了模糊支持向量机的理论框架,挑选出阿尔茨海默症数据集的11个特征变量中的前3个主成分和前2个主成分分别进行模型的训练。由于FSVM算法中的模糊因子能够识别一些特殊的样本点,故可通过赋予不同样本不同隶属度值来区分开信息量大的样本点与无用的噪声点。利用基于FSVM的模糊C均值聚类方法对阿尔茨海默症数据集的121个样本进行归类,得出较准确的分类结果,预测出的负类别准确率高达95.455%,但正类准确率稍低;(4)研究了半监督支持向量机的理论算法,具体分析模型中的各种函数与参数对分类结果的影响,并根据参数寻优找出最佳的学习模型。实验分析得出的分类准确率最高为94.118%且结果稳定,表明了S3VM方法能够综合利用标记样本和未标记样本的分布信息来提高模型的分类准确率。通过理论的探究与实验的验证可知,论文所研究的支持向量机中的第三种模型半监督支持向量机,相比于其他两种模型,在阿尔茨海默症数据的分析上具有更高和更稳定的分类准确率。表明该方法能够通过对脑功能数据进行分类来有效预测老年人是否患有阿尔茨海默症,以此更好地辅助医生进行AD的诊断与治疗。
[Abstract]:Alzheimer's disease (AD) is a chronic disease characterized by cognitive impairment. With the development of biomedicine, there are more and more data on Alzheimer's disease, but these data sets have the characteristics of high dimension, various forms and uneven distribution. How to make effective use of these complex data has become a hot issue in big data's time. Support vector machine (SVM) is a new tool for data mining based on statistical learning theory. However, the method can not recognize fuzzy labeled samples, nor can it use unlabeled samples, which leads to the deviation of model classification results. In order to deal effectively with complex data in Alzheimer's disease and not to waste a large number of valuable unlabeled samples, an improved support vector machine (SVM) algorithm is introduced. The fuzzy support vector machine (FSVM) and semi-supervised support vector machine (S3VM) are applied to the classification of Alzheimer's disease data. The accuracy of the classification results is observed by experiments. The main contents and results are as follows: (1) at first, the feature extraction method is used to process the data. In order to reduce the dimension of the data, 11 factor variables were extracted from 55 characteristic variables of 121 Alzheimer's data using principal component analysis. And these factor variables can basically represent all the information of the data. (2) the theoretical framework of support vector machine (SVM) is studied. For the kernel function and parameter problem in the SVM model, the classification experiments are carried out by setting different values. The degree of influence on classification accuracy was observed. The experimental results show that the SVM algorithm can effectively analyze Alzheimer's disease data, and the classification accuracy of test samples can reach 92.157. (3) the theoretical framework of fuzzy support vector machine (FSVM) is studied. The first three principal components and the first two principal components of 11 characteristic variables of Alzheimer's data set were selected for model training. Because the fuzzy factors in the FSVM algorithm can identify some special sample points, it is possible to distinguish the sample points with large amount of information from the useless noise points by giving different samples different membership values. A fuzzy C-means clustering method based on FSVM was used to classify 121 samples from Alzheimer's disease data set, and a more accurate classification result was obtained. The accuracy of negative class prediction is 95.455, but the accuracy of positive class is slightly lower. (4) the theoretical algorithm of semi-supervised support vector machine is studied, and the influence of various functions and parameters in the model on the classification results is analyzed. And find out the best learning model according to the parameter optimization. The experimental results show that the classification accuracy is 94.118% and the results are stable, which indicates that the S3VM method can improve the classification accuracy of the model by synthesizing the distribution information of labeled and unlabeled samples. Through theoretical research and experimental verification, we can see that the third model of support vector machine studied in this paper is semi-supervised support vector machine, compared with the other two models. It has higher and more stable classification accuracy in the analysis of Alzheimer's disease data. The results show that this method can effectively predict whether the elderly have Alzheimer's disease by classifying the brain function data, so as to better assist doctors in the diagnosis and treatment of AD.
【学位授予单位】：南阳师范学院
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R749.16

【相似文献】