Data Domain Description And Sample Reduction Research On Large-scale Datasets

Support vector machine (SVM) is a machine learning algorithm developed on statisticallearning theory. And it has many distinct advantages, such as achieving actual risk minimizationby looking for structural risk minimization, overcoming the local optimal solution, and having agood generalization and learning ability. Due to these benefits above, SVM is widely used in manyfields. However, SVM is put forward referring to small datasets. Therefore, during informationera when large-scale datasets often occur, how to improve SVM and transplant it onto large-scaledatasets becomes a research hot point in machine learning.The main task of this paper can be divided into two parts: the first part is the sampledescription of large-scale datasets. In terms of how to get rid of the traditional concept of singlesample point and adopt sample block to describe large-scale datasets and characteristicabstraction, simplified spheres description which applies to Gauss-distributed large-scale datasetsis put forward; the second part is sample reduction to large-scale datasets. In order to address theproblems such as reducing balanced classification error, improving the efficiency of classification,raising generalization and so on, a novel classification algorithm—Support Vector DomainClassification Hyper-plane (SVDCH),which combines the advantages of SVM and SVDD, isbrought up. And meanwhile, this paper also does some research on sample reduction approachesin SVM and SVDD. Combined with traditional sample reduction and based on SVDCH, Reduced-SVDCH is developed. And feasibility on big datasets is proved by experiments.