An improved LSSVM based on multi-scale wavelet kernel function

This work presents a methodology for the problem of the original Least Square Support Vector Machine (LSSVM) algorithm could not reach desired precision in multi-scale regression. In this methodology, we choose Mexican hat wavelet function as the kernel function of LSSVM firstly, and then the global optimum of the multi-scale regression modeling problem can be obtained by solving a quadratic programming problem. The regression model can effectively approximate multi-scale signals. The effectiveness of the proposed algorithm is validated by computer simulation results.