面向高维数据挖掘的特征选择方法研究-计算机软件与理论专业毕业论文.docxVIP

面向高维数据挖掘的特征选择方法研究 面向高维数据挖掘的特征选择方法研究 Titie：The Research on Feature Selection MethDds iIl Hi曲Dimensional Data Mining Major：Computer Software and Theory Name：Sun Jinghan Supervisor：Yin Ban Abstract Data mining is one of the frontiers of research in the fields of database and DSS．In practice，the hi。曲dimensional data is frequently used，as it becomes mainstream in real·life data increasingly,the research on high dimensional data mining is more and more important．However,mining in hi曲dimensional data is extremely difficult due to some particular characteristics of its own．Therefore，specific methods must be adopted to solve these problems． This paper starts with the concept of data mining and characteristics of hi【曲 dimensional data，regards feature selection methods in hi曲dimensional data mining as core content．Feature selection methods applied in text data and gene expression data are discussed respectively． For the problems such as dimension curve，sparse vector and random selection of initial centers in the area of text clustering and k-means algorithm，we proposed an improved k-means algorithm for text clustering，which is improved by feature selection,sparse vector filtering，density and scattering—based initial centers searching． Via using dataset 20 NewsGroup for experiment，the results show that the proposed algorithm works better than standard k-means algorithm both in quality and stability． For gene expression data，we proposed a new feature selection method，which realizes the feature subset search by genetic algorithm；and the feature subset fitness is evaluated by the separability measure based on boundary points．The experimental results show that the proposed algorithm can find the feature subsets with good separability,which results in the low dimensional data and the good classification accuracy．Based on the research mentioned above，another feature selection method ————————————————塑鳖塑塑丝塑竺竺堑兰兰塑鎏竺!垄 ————————————————塑鳖塑塑丝塑竺竺堑兰兰塑鎏竺!垄 一一一 combing filter and