Table 21.
Ablation study results on the independent 15% test subsets of the PlantVillage tomato and kaggle tomato leaf disease datasets.
| Model variation | Acc. (%) PV | Prec. (%) PV | Rec. (%) PV | F1 (%) PV | AUC PV | Acc. (%) KG | Prec. (%) KG | Rec. (%) KG | F1 (%) KG | AUC KG |
|---|---|---|---|---|---|---|---|---|---|---|
| Baseline YOLOv8n-CLS | 97.50 | 97.30 | 97.10 | 97.20 | 0.970 | 96.80 | 96.40 | 96.20 | 96.30 | 0.968 |
| YOLOv8n-CLS + Spatial Attention | 98.20 | 98.10 | 97.80 | 97.90 | 0.980 | 97.30 | 97.50 | 97.20 | 97.30 | 0.976 |
| YOLOv8n-CLS + Lightweight Classification Framework | 98.50 | 98.40 | 98.20 | 98.30 | 0.990 | 97.80 | 98.00 | 97.70 | 97.80 | 0.986 |
| Attention-Enhanced YOLOv8n-CLS | 98.90 | 98.70 | 98.50 | 98.60 | 0.991 | 98.45 | 98.20 | 97.50 | 97.80 | 0.988 |
PV, PlantVillage Tomato Dataset; KG, Kaggle Tomato Leaf Disease Dataset. All values represent single performance estimates obtained from the predefined independent 15% test subset of each dataset. The independent test subsets were held out from model training, five-fold cross-validation, hyperparameter tuning, and model selection.