Open Access
Table 4.
Calibration and test of machine-learning methods.
| Method | Tuning parameter | Explored range | Best value | Error rate (%) |
|---|---|---|---|---|
| Logistic regression | ∅ | 30.27 | ||
| Linear discriminant analysis | ∅ | 30.43 | ||
| k nearest-neighbors | k | [3600, 180 000] | 5000 | 23.79 |
| One-layer neural network | Dropout | [0.1, 0.5] | 0.2 | 22.51 |
| Nodes in each layer | [16, 256] | 128 | – | |
| Three-layer neural network | Dropout | [0.1, 0.5] | 0.2 | 21.06 |
| Nodes in each layer | [16, 256] | 128 | – | |
| Tree boosting (XGBoost) | Number of trees (nround) | [100, 1000] | 400 | 20.98 |
| Depth of each tree (max_depth) | [4, 15] | 12 | – | |
| Learning rate (eta) | [0.01, 0.2] | 0.1 | – | |
Notes. The best value of each tuning parameter was found by comparing error rates on the validation catalog. The error rate given in the last column is computed on the test catalog.
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.