The traditional SVM is more sensitive to the noise and isolated points in training sample, and have lower modeling accuracy. In this paper, fuzzy set theory is introduced to least squares support vector regression, and then to establish a data domain description fuzzy least squares support vector machine regression. This method will sample mapped into a high dimensional space, and search a minimum enclosing sphere in high dimensional space. Meanwhile, according to the distance from sample to the center of the sphere, the size of fuzzy membership can be determined. A simulation experiment is provided to demonstrate that this algorithm can improve the accuracy of support vector machine regression. This model is applied to predict the concentration of glutamic acid bacteria fermentation process. Results we obtain in simulation show the effectiveness of the proposed approach.