Table 3.
Feature importance of each prediction.
| Model | LogisticRegression | Random Forest | XGBoost | XGBH |
|---|---|---|---|---|
| 1 | Ap_hi (0.1383 ± 0.0070) | Ap_hi (0.1326 ± 0.0028) | Ap_hi (0.13750 ± .0065) | Ap_hi (0.1406 ± 0.0047) |
| 2 | Weght (0.1218 ± 0.0056) | Chol (0.0302 ± 0.0052) | Chol (0.0321 ± 0.0050) | Chol (0.0358 ± 0.0034) |
| 3 | BMI (0.0473 ± 0.0030) | Age (0.0239 ± 0.0058) | Age (0.0268 ± 0.0030) | Age (0.0276 ± 0.0043) |
| 4 | Height (0.0434 ± 0.0050) | Active (0.0024 ± 0.0016) | Ap_lo (0.0059 ± 0.0007) | Ap_lo (0.0063 ± 0.0007) |
| 5 | Age (0.0319 ± 0.0079) | Ap_lo (0.0023 ± 0.0026) | BMI (0.0045 ± 0.0026) | BMI (0.0036 ± 0.0023) |
| 6 | Chol (0.0012 ± 0.0008) | Smoke (0.0008 ± 0.0017) | Active (0.0020 ± 0.0011) | Active (0.0034 ± 0.0017) |
| 7 | Smoke (0 ± 0.0000) | Gender (0.0002 ± 0.0043) | Height (0.0018 ± 0.0015) | Gender (0.0016 ± 0.0010) |
| 8 | Alco ( 0.0000 ± 0.0001) | Alco ( 0.0000 ± 0.0008) | Gender (0.0014 ± 0.0009) | Gluc (0.0015 ± 0.0011) |
| 9 | Active ( 0.0001 ± 0.0004) | Gluc ( 0.0012 ± 0.0014) | Smoke (0.0012 ± 0.0009) | Smoke (0.0012 ± 0.0005) |
| 10 | Gluc ( 0.0001 ± 0.0003) | BMI ( 0.0092 ± 0.0022) | Weight (0.0010 ± 0.0024) | Weight (0.0010 ± 0.0007) |
| 11 | Gender ( 0.0002 ± 0.0003) | Height ( 0.0100 ± 0.0049) | Aclo (0.0003 ± 0.0003) | Height (0.0010 ± 0.0012) |
| 12 | Ap_lo ( 0.0003 ± 0.0006) | Weight ( 0.0125 ± 0.0023) | Gluc (0.0002 ± 0.0005) | Aclo (0.0004 ± 0.0006) |
The value of weight plus or minus represents the half of 95% confidence interval length. The larger the feature weight, the greater the feature predictability, and the negative weight represents an inhibitory effect on prediction.