Considering the existence of significant difference among the training samples

a weighted solution path algorithm for support vector regression is proposed in which the error penalty parameter on each training sample is weighted differently. The weighting coefficients are obtained by using piecewise linear interpolation based on the difference of samples' importance. Then the solution path is adjusted by the resulting weighting coefficients

such that the effects of different samples in regression model are changed correspondingly. The experiments on the prediction of response data of cylinder structure under two different boundary conditions in time frequency domain

where the importance of each training sample is characterized by the geometrical distance to the test sample

show that the weighted solution path algorithm for support vector regression can reduce prediction errors of displacement response on different evaluation indexes

and that the generalization of regression model is improved. It is noted that the main idea of the proposed algorithm can also be applied to other solution path algorithms

such as λ-path and ν-SVR path algorithms.