Abstract: To improve the classification performance of unbalanced datasets, the nature characteristics of sparse least squares support vector machines (LS-SVM) was analyzed and an algorithm based on center distance ratio for the unbalanced samples was proposed. Meanwhile, the problem of sparseness lacking in the least squares support vector machines was solved by pruning the boundary samples. The new algorithm was tested on the UCI datasets. The results indicate that this method can effectively improve the classification accuracy of LS-SVM for the unbalanced samples, the proposed algorithm can properly obtain the sparse solutions without affecting the capacity of classification, and the speed of classifiers is also improved.