Table 3.
Model experimental results with 95% confidence intervals.
| Methods | R 2 | EVS | MAE |
|---|---|---|---|
| Prediction results of palpebral conjunctiva images artificially selected by feature engineering method | |||
| Decision tree | 0.262 (0.242, 0.283) | 0.267 (0.247, 0.287) | 2.054 (2.028, 2.080) |
| Linear regression | 0.300 (0.288, 0.312) | 0.304 (0.292, 0.315) | 1.995 (1.979, 2.010) |
| SVM | 0.267 (0.248, 0.286) | 0.270 (0.252, 0.289) | 2.042 (2.019,2.064) |
| K-nearest neighbor regression | 0.249 (0.230, 0.268) | 0.251 (0.233, 0.27) | 2.057 (2.036,2.078) |
| Random forest regression | 0.285 (0.270, 0.299) | 0.287 (0.273, 0.302) | 2.028 (2.010, 2.047) |
| Boosting tree regression | 0.296 (0.283, 0.308) | 0.298 (0.287, 0.31) | 2.012 (1.995,2.029) |
| Prediction results of original eye images base on feature engineering | |||
| Decision tree | −0.013 (−0.032, 0.006) | −0.001 (−0.021, 0.018) | 2.425 (2.404,2.447) |
| Linear regression | −0.077 (−0.113, −0.041) | −0.064 (−0.101, 0.027) | 2.462 (2.427,2.496) |
| SVM | −0.052 (−0.081, −0.024) | −0.039 (−0.068, −0.01) | 2.428 (2.401,2.455) |
| K-nearest neighbor regression | −0.021 (−0.040, −0.001) | −0.006 (−0.026, 0.014) | 2.401 (2.382, 2.420) |
| Random forest regression | −0.019 (−0.041, 0.003) | −0.004 (−0.027, 0.019) | 2.421 (2.397,2.444) |
| Boosting tree regression | −0.053 (−0.083, −0.024) | −0.037 (−0.067, 0.007) | 2.442 (2.410, 2.473) |
| Prediction results of deep CNNs based on a priori causal knowledge | |||
| BCNN | 0.447 (0.446, 0.447) | 0.452 (0.451, 0.453) | 1.812 (1.812, 1.813) |
| mobilev2 | 0.447 (0.445,0.450) | 0.462 (0.459, 0.466) | 1.822 (1.819, 1.826) |
| Shufflenetv2 | 0.321 (0.319,0.323) | 0.357 (0.352, 0.364) | 1.983 (1.976, 1.991) |
| Squeezenet | 0.498 (0.495,0.502) | 0.511 (0.508, 0.514) | 1.688 (1.685, 1.693) |
| Resnet_cbam | 0.463 (0.461,0.466) | 0.463 (0.461,0.466) | 1.719 (1.714, 1.725) |
| mobilenetv3+SE | 0.512 (0.499,0.517) | 0.535(0.515, 0.542) | 1.521 (1.481, 1.574) |