Table 5.
Sensitivity, PPV, F1 score, and NPV average across all trials for individual classes (containing any of the 3 concepts or none) on bootstrapped external validation for the rule-based model, Machine Learning model, Bi-LSTM_simple model, and Bi-LSTM_dropout model, where 0 = no concept and 1 = concept present
| Sensitivity | PPV | F1 score | NPV | ||
|---|---|---|---|---|---|
| Rule-based model | Overall | 0.73 | 0.87 | 0.74 | 0.73 |
| 0 | 0.99 | 0.75 | 0.86 | 0.99 | |
| 1 | 0.46 | 0.98 | 0.63 | 0.65 | |
| ML model | Overall | 0.50 | 0.56 | 0.39 | 0.50 |
| 0 | 1.0 | 0.62 | 0.77 | 1 | |
| 1 | 0.004 | 0.50 | 0.009 | 0.50 | |
| Bi-LSTM_simple model | Overall | 0.55 | 0.57 | 0.54 | 0.56 |
| 0 | 0.83 | 0.66 | 0.73 | 0.63 | |
| 1 | 0.28 | 0.49 | 0.34 | 0.54 | |
| Bi-LSTM_dropout model | Overall | 0.50 | 0.54 | 0.39 | 0.50 |
| 0 | 0.99 | 0.62 | 0.77 | 0.65 | |
| 1 | 0.006 | 0.47 | 0.01 | 0.50 |