Application On Semi-supervised Support Vector Machine To The Data Analysis Of Alzheimer’s Disease

Alzheimer’s disease, a kind of chronic disease, whose main clinical feature is cognitive impairment, is also a high incidence of old-age disease commonly. With development of biomedical science, more and more medical data have been produced on the study of Alzheimer’s disease, but these data sets have characteristic of high dimension, a variety of forms and uneven distribution. How to use these complex data effectively becomes a hot issue that the era of big data would discuss.Support vector machine(SVM) comes into being based on the statistical learning theory. It is a new kind of data mining tool using optimization method. However this method is unable to identify the fuzzy labeled samples or use lots of unlabeled samples, so that model classification results appear bias. In order to deal with complex data of Alzheimer’s disease effectively, and not waste a lot of valuable unlabeled samples data, we introduce two improved algorithms based on support vector machine, which are fuzzy support vector machine(FSVM) and semi-supervised support vector machine(S3VM), then apply the two methods to the classification technology of dementia data, and observe the accuracy of the classifying results by experiments. The research content and the results are as follows:(I)First of all, we utilize the feature extraction method to process data in the early period. To reduce the dimension of data, we use principal component analysis method to extract out eleven factors from fifty-five characteristic variables of one hundred and twenty-one Alzheimer’s samples data. And these factor variables could represent all the information in the data basically.(II)We research the support vector machine theory in detail. For the problems of kernel functions and parameters, we do the classification experiment by setting its different values and observe its effect on the accuracy of classification. The experimental results show that the support vector machine algorithm could be used to analyze the Alzheimer’s data effectively, and the classification accuracy of test samples is up to 92.157%.(III)We study the theory framework of fuzzy support vector machine, and select the former three or two principal components of Alzheimer’s data’s eleven characteristic variables to train classification model respectively. Due to fuzzy factor of the model could identify some special sample points, so we could give different samples different membership values to separate informative sample points and useless noise sample points. Then, we use the fuzzy c-means clustering method based on fuzzy support vector machine to classify the one hundred and twenty-one samples of Alzheimer’s disease, and acquire more accurate classification results. Finally, the accuracy of negative category is predicted as high as ninety-five point four five five percent, but positive category accuracy is lower.(IV)We research the theory algorithms of semi-supervised support vector machine, and analyze the influence of various functions and parameters in model on the classification results in detail, then find the best learning model according to the parameter optimization. By experiment analysis, we come to the conclusion that the highest classification accuracy is 94.118% and also is stable. The results indicate that semi-supervised support vector machine method can improve the classifition accuracy of models utilizing labeled samples and unlabeled samples distribution information synthetically.Through study of theory and experiment validation, we know that the third model named semi-supervised support vector machine of support vector machine has higher and more feasibility classification accuracy comparing with other two kinds of models. The results show that this method could predict dementia patients effectively through classifying brain function data, so that it could assist doctors to diagnose and treat Alzheimer’s disease.