Table 3 Effect of various synthetic data generation and classification techniques on F1-score improvements and reductions in the COVID-19, kidney and dengue datasets.
Data | Highest F1-score upgrade | Combined method | Model | Synthetic data | Highest F1-score down-grade | Combined method downgrade | Model | Synthetic data |
|---|---|---|---|---|---|---|---|---|
COVID | + 0.15 | ADASYN + DCTGAN + ResNet(Poposed) | TabNet | + 50% | − 0.30 | SMOTE + DCTGAN+ ResNet | KNN | Only-synth |
COVID | + 0.12 | NC + DCTGAN + ResNet | XGB | + 25% | − 0.22 | BS + Sep. + DCTGAN+ ResNet | TabNet | + 200% |
COVID | + 0.10 | SMOTE + DCTGAN + ResNet | RF | + 75% | − 0.18 | NC + DCTGAN+ ResNet | RF | + 100% |
Kidney | + 0.20 | ADASYN + DCTGAN + ResNet(proposed) | TabNet | + 100% | − 0.25 | NC + Sep. + DCTGAN+ ResNet | KNN | Only-synth |
Kidney | + 0.05 | BS + DCTGAN + ResNet | RF | + 200% | − 0.20 | ADASYN + DCTGAN+ ResNet | TabNet | + 400% |
Kidney | + 0.03 | NC + DCTGAN + ResNet | XGB | + 50% | − 0.15 | SMOTE + DCTGAN+ ResNet | KNN | + 100% |
Dengue | + 0.18 | SMOTE + DCTGAN + ResNet(proposed) | TabNet | + 150% | − 0.24 | NC + DCTGAN+ ResNet | XGB | + 150% |
Dengue | + 0.11 | NC + DCTGAN + ResNet | KNN | + 75% | − 0.19 | SMOTE + DCTGAN+ ResNet | RF | + 300% |
Dengue | + 0.08 | BS + DCTGAN + ResNet | RF | + 100% | − 0.17 | BS + Sep. + DCTGAN + ResNet | KNN | + 200% |
