Weighted solution path algorithm of support vector regression for abnormal data

In the solution path algorithm of support vector regression, the penalty for violation of the required error is considered equally for every training sample, which means every training sample affects the generalization ability equally. Considering the existing of abnormal samples among the training data, for example noises with different variances, the weighted solution path algorithm of support vector regression is proposed. To reduce the negative effect of abnormal samples, different weighting coefficients are set on the error penalty parameter of corresponding samples. The whole solution path can be adjusted correspondingly. So the effects of abnormal samples on regression model have been reduced by setting lower coefficient. Experiments demonstrate that accuracy of prediction and the generalization of regression model can be improved.