Table 6.
Accuracy and F1 score of the reduced MEPSa–trained model for the MEPS and PatientsLikeMe testing datasets. For comparison, the second row category of the table repeats the results previously obtained for the original model with all predictors from Table 4.
| Model | MEPS dataset | PatientsLikeMe dataset | |
| Reduced model (without daily intake and out-of-pocket expense) | |||
|
|
Accuracy (%) | 73.2 | 67.8 |
|
|
F1 score | 73.9 | 79.7 |
| Original model (with daily intake and >out-of-pocket expense) | |||
|
|
Accuracy (%) | 76.2 | 54.9 |
|
|
F1 score | 77.1 | 65.5 |
aMEPS: Medical Expenditure Panel Survey.