Table 3.
Test Results of Best-Corrected Visual Acuity Regression for Each Model
| Model | R2 (95% CI) | RMSE (95% CI) | MAE (95% CI) |
|---|---|---|---|
| Benchmark models | |||
| Linear | 0.306 (−0.0702, 0.584) | 12.8 (8.85, 16.9) | 9.46 (7.08, 12.1) |
| Random forest | 0.0825 (−0.499, 0.513) | 14.7 (9.71, 19.4) | 10.4 (7.29, 13.7) |
| XGBoost | 0.297 (−0.143, 0.606) | 12.8 (8.5, 17.3) | 9.46 (6.95, 12.2) |
| Deep neural network | 0.0786 (−0.246, 0.351) | 14.7 (10.3, 19.0) | 10.9 (8.12, 14.2) |
| Model stacking | |||
| Linear | 0.308 (−0.0261, 0.583) | 12.7 (8.61, 16.9) | 9.02 (6.39, 11.9) |
| Random forest | 0.147 (−0.413, 0.566) | 14.1 (9.26, 18.5) | 10.0 (7.21, 13.2) |
| XGBoost | 0.292 (−0.0708, 0.561) | 12.9 (8.73, 17.0) | 9.29 (6.68, 12.1) |
| Model averaging | |||
| Linear | 0.270 (0.0358, 0.479) | 13.1 (9.20, 16.8) | 9.80 (7.36, 12.6) |
| Random forest | 0.201 (−0.107, 0.456) | 13.7 (9.44, 17.8) | 10.1 (7.46. 13.0) |
| XGBoost | 0.273 (0.00396, 0.478) | 13.0 (9.10, 16.7) | 9.80 (7.37, 12.5) |
CI = confidence interval; MAE = mean absolute error; R2 = coefficient of determination; RMSE = root mean squared error; XGBoost = extreme gradient boosting.