|
系统科学与数学 2010
LINEAR ε-SUPPORT VECTOR REGRESSION DATA PERTURBATION ANALYSIS
|
Abstract:
Support Vector Regression(SVR) is one of the important method for regression problem. As linear regression problem has a variety of applications in the real life, linear $\varepsilon$-Support Vector Regression is an object of study. Because the solution of the primal problem is our final foucus on, the is to eventual aim establish the sensitivity analysis theory for the primal problem of linear $\varepsilon$-SVR, which deals with the case of some inputs of training data perturbation problem. The expression of solution to the partial derivative via the perturbed data under some conditions, and the sensitivity analysis theorem of the primal problem of $\varepsilon$-SVR is given. Finally, some simple simulation results show that the theorem is effective.