Estimating Precisions for Multiple Binary Classifiers Under Limited Samples