Research On Algorithms And Applications Of Support Vector Regression

Support Vector Machine (SVM) as a new approach of machine learning in recent years, has made a rapid progress in theoretic research and practical application. Support vector machine has successfully resolved regression (time series analysis) and pattern recognition (classification, discriminate analyze), etc with its calculation advantages and applicable feasibility, which could be further applied in the fields such as forecasting and multiple criteria evaluation.Support vector machine was proposed for pattern recognition at first, but as the introduction of e-insensitive loss function in recent years, it has extended over the area of Nonlinear Regression. And it represents a good learning capability. Support vector machine become one of best ways of dealing with regression.Sequential Minimal Optimization (SMO) effectively solves SVM's problems, such as complex realization, low efficiency, etc. Moreover, excellent precision and calculation efficiency could be achieved by the execution of SMO in case of huge data processing. At present, SMO has become the mainstream resolution of SVM applied in huge data case.This paper makes detailed instruction on SMO of SVR and issues revised arithmetic for parameters choosing problems. There are five chapters in this paper. Chapter 1 generally discusses SVM theory, its development and present research situation and the problems to resolved in this research; Then, this paper expounds Support vector classification and regression theories from both linear and non-linear situations in Chapter 2; Chapter 3 introduces original SMO arithmetic; Revised SMO arithmetic is issued in Chapter 4 for parameter choose problem in SVR, meanwhile its feasibility proved by data examination in which high efficiency of revised arithmetic is validated by compared with known method. In last Chapter, the revised arithmetic is applied in distribution centers site selection and foetus weight regression which makes importantly practical sense, and method for choosing parameters of SVR is outlined as well.