Support Vector Regression And Its Applications

Support Vector Machine (SVM) is a novel powerful machine learning method rooted in the framework of Statistical Learning Theory (SLT). SVM has become a new active area in the field of artificial intelligent and machine learning because of its excellent learning performance. SVM consists essentially of Support Vector Classification and Support Vector Regression. Of them, the theory and applications of SVC have gradually come to maturity, while the research of SVR is weak either in scope or in depth.Different loss functions will cause different empirical risks, and also lead to different SVRs. At present, SVRs using linear loss function are more concentrated than those using other linear loss functions. The main works in this thesis are as follows:(1) Three kinds of SVRs using ε -insensitive loss function, quadratic ε -non loss function and Huber loss function are elaborated, compared and a general conclusion is reached in addition.(2) SVRs using ε -insensitive loss function are not always optimal for the particular problems. Therefore, a new loss function combining the merits of three different kinds of loss function is put forward for the new SVR algorithm—RSVR.(3) Model selection is an important research direction in SVM. A new extra parameter c is drawn into RSVR algorithm. Parameters c and C have a direct impact on the generalization performance of the new algorithm. Based on the experimental results, this thesis presents an integrated algorithm named PSO-RSVR using Particle Swarm Optimization for model selection.(4) Applications of SVRs.(a) In stock markets, because of its high benefits and risks, the price forecasting is an active study field for both investors and scholars. However, for the difficulty of its highly nonlinear, numerous methods applied are not agreeable. On the basis of the conclusions above, PSO-RSVR is proposed for stock price forecasting with effective results.(b) Temperature compensation has been a hot research issue. For the output of the pressure sensor is easily affected by temperature, two methods of...