Learning Curves. The learning curves provide an analysis of system performance relative to dataset size. The dotted line shows the addition of GE training data to ID training data. The x-axis is binary logarithmic, and the training corpus size roughly doubles between most points in the curves (2, 4, 8, 16, 32, 64 and 100%). Thus, a linear growth in F-score indicates a need for a corresponding exponential increase in dataset size.
Björne et al. BMC Bioinformatics 2012 13(Suppl 11):S4 doi:10.1186/1471-2105-13-S11-S4