Table 1. Barzooka screening tool performance in an internal validation set.
Class | Manually labeled cases | False positives | Precision | Recall | F1 score |
---|---|---|---|---|---|
Bar graph of counts or proportions (appropriate) | 407 | 65 | 0.84 | 0.86 | 0.85 |
Bar graph of continuous data (inappropriate) | 671 | 35 | 0.95 | 0.91 | 0.93 |
Bar graph with dot plot | 149 | 10 | 0.93 | 0.91 | 0.92 |
Dot plot | 393 | 33 | 0.91 | 0.85 | 0.88 |
Box plot | 368 | 29 | 0.92 | 0.88 | 0.90 |
Violin plot | 340 | 13 | 0.96 | 0.94 | 0.95 |
Histogram | 238 | 32 | 0.86 | 0.86 | 0.83 |
Flow chart | 276 | 32 | 0.89 | 0.91 | 0.90 |
Pie chart | 160 | 5 | 0.97 | 0.92 | 0.94 |
The internal validation set included 10% (n=3812 pages) of the pages classified to train the algorithm.