Table 1.
Performance of the models for the first task of article screening using the original datasetA
| srBERTmy250K | srBERTmix | Original BERT | K-neighbors | SVC | DecisionTree | RandomForest | Adaboost | MultinomialNB | |
|---|---|---|---|---|---|---|---|---|---|
| AUC | 76.785 | 50.000 | 50.685 | 57.985 | 50.000 | 57.449 | 53.650 | 55.097 | 50.000 |
| Accuracy | 94.353 | 89.945 | 90.083 | 90.083 | 89.945 | 89.118 | 89.945 | 90.358 | 89.945 |
| Precision | 83.333 | 0.000 | 100.000 | 52.000 | 0.000 | 40.620 | 50.000 | 61.538 | 0.000 |
| Recall | 54.795 | 0.000 | 13.60 | 17.808 | 0.000 | 17.808 | 8.219 | 10.959 | 0.000 |
| F1 | 66.116 | 0.000 | 26.84 | 26.531 | 0.000 | 24.762 | 14.118 | 18.605 | 0.000 |
SR systematic review, BERT bidirectional encoder representations from transformers, srBERTmy250K srBERTmy model trained for 250 K steps, AUC area under the curve, SVC support vector classification, MultinomialNB multinomial naive Bayes model