Fig. 1.

Gain in classification accuracy (relative to the “majority vote” baseline classifier) plotted against runtime. Each point represents one of 25 datasets, with size proportional to the input dimension. Note the x-axis is drawn on the log-scale. CoDaCoRe (with balances) is the only method that scales effectively to our larger datasets, while consistently achieving high predictive accuracy. Moreover, its performance is broadly consistent across smaller and larger datasets