Table 2.
Common methods in each machine learning category
| Machine learning type | Methods | Number of articles |
|---|---|---|
| Traditional supervised learning | Random forest | 14 |
| Logistic regression | 11 | |
| Support vector machine (SVM) | 11 | |
| L1-penalized logistic regression | 8 | |
| Decision trees | 4 | |
| Extreme gradient boosting (XGBoost) | 4 | |
| Naive Bayes | 3 | |
| Deep supervised learning | Recurrent neural networks (RNNs) and variants | 19 |
| Convolutional neural networks (CNNs) and variants | 11 | |
| BERT and variants | 7 | |
| Feed-forward neural networks (FFNNs) | 3 | |
| Weakly supervised learning | PheNorm74 | 3 |
| MAP75 | 2 | |
| Random forest (with silver-standard labels) | 2 | |
| Unsupervised learning | Latent Dirichlet Allocation (LDA) | 5 |
| K-means | 4 | |
| UPGMA (Unweighted Pair Group Method with Arithmetic mean) hierarchical clustering | 2 |
Note: A method is presented if it appeared in more than 1 article. Several papers used more than 1 method. The table does not include any semi-supervised methods as each article used a distinct method. Semi-supervised methods are presented in Table 3.