Table 2:

Average percent enrichment at a ranking of 10% for classifiers. We consider both the performance of classifiers trained on the fragments selected in a single run (〈ϵ(10%)〉_sing) and the performance of classifiers trained on the top nine fragments returned in at least 5 runs of the algorithm (〈ϵ(10%)〉_top9). We report the average performance as the mean of the 28 means of the 5 classifiers along with the standard error in the mean.

Class	〈ϵ(10%)〉sing (%)	〈ϵ(10%)〉top9 (%)
0	65 ± 1	82.0 ± 0.5
1	32 ± 8	154 ± 2
2	134 ± 4	133 ± 2
3	126 ± 3	173 ± 3
4	107 ± 3	141 ± 2