Skip to main content

Table 3 Statistics about the comparative performance of the base k-NN classifiers and their label similarity-incorporated versions on small classes (size ≤ 30), measured in terms of the number of classes for which AUC scores are improved by the latter over the former, and the average and maximum improvement in AUC scores over all classes.

From: Incorporating functional inter-relationships into protein function prediction algorithms

Dataset # Small classes # Classes improved Average improvement over all small classes Maximum improvement
Mnaimneh 47 27 0.0358 (6.24%) 0.1882 (39.92%)
Rosetta 48 21 0.0225 (3.82%) 0.2091 (38.66%)
Krogan 40 14 0.0129 (1.89%) 0.1982 (31.82%)
Combined 48 28 0.0197 (2.72%) 0.1129 (20.39%)