Application Research Of Grey Support Vector Machines On Small Sample Set

Grey System Theory (GST) studies on the indeterminate system with "a few samples" and" poor" information, which is in the situation of "part of information known, part of information unknown" .By generating and developing the "part of information known" , GST can help us understand and recognize the real world, and help us rightly master and describe the operational behavior and evolutional law of the investigated ystem. Grey Dynamic Modeling Technique is the core of GST, and is also the bridge between the GST and practice.Support vector machines (SVM) is a novel and powerful machine learning approach developed in the framework of statistical learning theory, which bases on the VC theory and the principle of structural risk minimization. It always performs well in many prctical applications with high generalization because of its better traditional learning approaches,such as Nearal Network, SVM holds the advantages of good generalization, being insensitive to high dimension data and convergence to global optimum, so it solves the intractable problems of the former, such as over-learning,local minima, dimension curese etc.The main works of this paper include the following three parts:1. A admixture algorithm is presented base on grey relational analysis and support vector machines. Pretreament module which grey relational analysis attribution reduction algorithm course endow different weight to each influencing fators, At last the predictive performance is checked.2. Introducing Grey prediction model GM(1,1), and analyzing the mechnism of GM(1, 1) to identify the impact of model accuracy of a variety of factors(Background Value, Initial Value, Smoothness),then three prediction models are presented based on each impact factors, that is, Background value prediction model(BGM model);, Initial value prediction model(IGM model) and Smoothness prediction model(SGM model). 3. Treating the prediction data of BGM model, IGM model, SGM model as input factors, the actual data as output factor, At last the predictive performance is checked.