Table A.1.
Set of best values obtained from a grid search-based optimization of the five hyperparameters that typically have the most influence in the performance of an RF classifier.
| RF classifier | n_estimators | max_depth | min_samples_split |
|---|---|---|---|
| r, 54 visits | 500 | 10 | 10 |
| r, 33 visits | 100 | 10 | 10 |
| g, 33 visits | 300 | 20 | 2 |
| rg, 33 visits | 500 | 10 | 10 |
| rg + bivar., 33 visits | 500 | 10 | 10 |
| (g − r)feat, 33 visits | 100 | 10 | 10 |
| (g − r)mag, 33 visits | 500 | 10 | 10 |
| r, 33 real, 0 synthetic | 300 | 10 | 2 |
| r, 29 real, 4 synthetic | 300 | 10 | 10 |
| r, 25 real, 8 synthetic | 100 | 10 | 10 |
| r, 21 real, 12 synthetic | 100 | 10 | 2 |
| r, 17 real, 16 synthetic | 300 | 10 | 10 |
| rg, 25feat, 33 visits | 300 | 10 | 10 |
| rg, 9feat, 33 visits | 100 | 20 | 2 |
| rg, 8feat, 33 visits | 100 | 20 | 10 |
| rg, 7feat, 33 visits | 100 | 10 | 2 |
Notes. The optimized hyperparameters are n_estimators, max_depth, min_samples_split, min_samples_leaf, and max_features; these were introduced in Sect. 4). The first column in the table lists the various classifiers tested in this work, whose performance metrics are reported in Table 4. We do not include a column for the last two hyperparameters since our tests always returned the same best values min_samples_leaf = 4 and max_features = sqrt, the only exception being the ks8 classifier, for which max_features = log2.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.