Table 1
Metrics for our models.
| Model | Accuracy | Precision | Recall | F-score |
|---|---|---|---|---|
| Full | 97.1% | 82.4% | 73.3% | 77.6% |
| No fit | 95.0% | 69.4% | 48.7% | 57.2% |
| Cutoff | 87.0% | 33.4% | 89.5% | 48.6% |
| Majority | 93.1% | NA | 0.0% | NA |
| Important feat. | 96.2% | 77.8% | 70.9% | 74.2% |
| SMOTE | 95.6% | 69.2% | 77.9% | 73.3% |
Notes. Performance metrics for the different models considered. The full model is trained on all available features. The restricted model (“No fit”) is trained on a subset of model-independent features. The cutoff model relies on a flare template fit. The majority model always predicts the majority class. The “Important feat” model uses only the most PFI-important features. The SMOTE model is trained on a dataset augmented using the SMOTE algorithm.
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.