According to the basic principle of support vector regression algorithm, for the difference of features` correlative degree to the regression, the affect of parameters to the performance of forecast and taking into account the significance of weighted feature after normalization, least squares support vector regression machine (LS-SVR) based on weighted feature is proposed in this paper, in which, least squares support vector regression algorithm is used to reduce the number of parameters and exponentially weighted feature is used to improve prediction accuracy, the weighting coefficients are determined by the grey correlation degree approach. In the meantime, the effectiveness of the algorithm is demonstrated in forecasting the actual stock price. The experimental results show that it is superior to classical support vector machine and can significantly improve the predictive ability.