A Study Of Wavelet Kernel In Support Vector Regression
Lefei Zhang, Fengqing Han · Intelligent Automation & Soft Computing · 2010
Abstract A new kernel function is proposed for support vector regression (SVR). A one-to-one mapping is adopted for dimensionality reduction and then continuous wavelettransform is utilized to construct the nonlinear mapping φ(x) from the input space Sto the feature space. So we call it continuous wavelet kernel function (CWKF). This wavelet kernel is not translation invariant kernel, instead inner product kernel and need not parameter selecting. The quadratic program of support vector regression has feasible solution if we use CWKF. Numerical experiments demonstrate the effectiveness of this method.