Abstract: Data classification is the basis of pattern recognition and fault diagnosis technologies. In practical applications, the conventional classification algorithms have some limitations due to nonlinear, inducednoise, and imbalances in the sample collection. In this paper, weighted least squares support vector machine is utilized in the problem of classification. By means of K-clustering algorithm, the intrinsic relationship between samples is analyzed and the weight coefficients are determined such that the impact of noise can be reduced and the number imbalance of different types of samples can be compensated. Moreover, the training time may be reduced and the accuracy of classification may be improved. Finally，an example of multi-class classification for image segmentation is given to illustrate the efficiency of the proposed approach.