Supplementary Material S1
Manuscript: Classical Machine Learning versus Transfer Learning for Tomato Leaf Disease Classification: A Reproducible Comparative Framework

This archive contains reproducibility-oriented outputs from the final experimental evaluation.

Files included:

1. predicciones_test.csv
   Per-image predictions for the 2,736 independent test images.
   Columns: file, y_true, KNN, SVM, Random_Forest, MobileNetV2.
   Class labels are encoded using the same deterministic label order employed in the manuscript.

2. metricas_test_finales.csv
   Final test-set accuracy, macro-precision, macro-recall, and macro-F1 for KNN, SVM-RBF, Random Forest, and MobileNetV2.

3. accuracy_wilson_ci.csv
   Test-set accuracy estimates and 95% Wilson confidence intervals for all four classifiers.

4. bootstrap_macro_metrics_ci.csv
   Stratified nonparametric bootstrap estimates and percentile 95% confidence intervals for accuracy, macro-precision, macro-recall, and macro-F1.
   Number of bootstrap resamples: 2,000.
   Random seed: 42.

5. mcnemar_pairwise_holm.csv
   Exact two-sided McNemar pairwise comparisons based on classification correctness for the same independent test images.
   Holm adjustment was applied across the six pairwise comparisons.

6. mobilenetv2_training_history.csv
   Training and validation accuracy/loss trajectories for the corrected two-stage MobileNetV2 transfer-learning experiment, including learning rate and training phase.

Notes:
- The original PlantVillage images are not redistributed in this supplementary archive.
- Data augmentation was applied only to the training subset.
- Validation and test subsets were not augmented.
- The independent test set contained 2,736 images.
- These files correspond to the final results reported in the submitted manuscript.

Suggested manuscript description:
“Supplementary Material S1 contains test-set predictions, final performance metrics, Wilson confidence intervals, stratified bootstrap confidence intervals, exact McNemar pairwise comparisons with Holm correction, and the MobileNetV2 training history.”
