Table 3.
Metrics Used to Evaluate Machine Learning Algorithms Ability by CPT Codes.
| Long Short Term Memory Network | Random Forest Machine Learning | |
|---|---|---|
| Accuracy to senior billing coder | 59.0% | 87.0% |
| Accuracy by CPT code | ||
| 22 856 | — | 98.4% |
| 20 931 | — | 96.0% |
| 22 858 | — | 98.4% |
| 20 931 | 89.0% | — |
| 69 990 | 84.0% | — |
| 22 551 | 75.0% | — |
| Area under the receiver operating curve by CPT code | ||
| 22 856 | — | .99 |
| 22 858 | — | .99 |
| 69 990 | .88 | .96 |
| 20 931 | .92 | — |
| 22 554 | .89 | — |
| Area under the precision-recall curve by CPT code | ||
| 20 931 | — | .96 |
| 22 858 | — | .93 |
| 22 856 | — | .94 |
| 22 551 | .89 | — |
| 22 845 | .79 | — |
| 20 931 | .95 | — |