Table 3. Frequencies (%) of observed and predicted pollen risk grade levels by the regression, SVR and DNN models at the 9 sites in Korea from 2015 to 2016.
| Observed risk grade | Predicted risk grade | |||||
|---|---|---|---|---|---|---|
| Model | Mild (%) | Moderate (%) | Severe (%) | Extreme (%) | Total | |
| Mild | Regression | 1,819 (89.1) | 87 (4.3) | 86 (4.2) | 49 (2.4) | 2,041 |
| SVR | 1,543 (75.6) | 423 (20.7) | 75 (3.7) | 0 (0.0) | ||
| DNN | 1,697 (83.1) | 238 (11.7) | 94 (4.6) | 12 (0.6) | ||
| Moderate | Regression | 51 (58.6) | 6 (6.9) | 26 (29.9) | 4 (4.6) | 87 |
| SVR | 6 (6.9) | 61 (70.1) | 20 (23.0) | 0 (0.0) | ||
| DNN | 16 (18.4) | 33 (37.9) | 34 (39.1) | 4 (4.6) | ||
| Severe | Regression | 30 (65.2) | 6 (13.0) | 6 (13.0) | 4 (8.7) | 46 |
| SVR | 3 (6.5) | 33 (71.7) | 10 (21.7) | 0 (0.0) | ||
| DNN | 6 (13.0) | 12 (26.1) | 26 (56.5) | 2 (4.3) | ||
| Extreme | Regression | 8 (36.4) | 5 (22.7) | 8 (36.4) | 1 (4.5) | 22 |
| SVR | 4 (18.2) | 12 (54.5) | 6 (27.3) | 0 (0.0) | ||
| DNN | 4 (18.2) | 6 (27.3) | 10 (45.5) | 2 (9.1) | ||
SVR, support vector regression; DNN, deep neural network.