Update README.md

README.md

@@ -72,8 +72,29 @@ AutoGluon automatically searched across the following dimensions:
 ---
 
 ## Results
-- **Best model (AutoGluon selected):** *efficientnet_b0*  
-- **Validation Accuracy:** *0.96875*   
+
+### Best Architecture
+- AutoGluon selected **EfficientNet-B0** as the best performing backbone in terms of validation accuracy.  
+- Other backbones tested included **ResNet18** and **MobileNetV3-Small**, which had slightly lower validation accuracy.
+
+### Best Hyperparameters
+- Optimizer: AdamW  
+- Learning rate: ~0.001 (exact value depends on AutoGluon’s internal selection)  
+- Weight decay: ~1e-4  
+- Regularization: implicit (from backbone architecture)  
+- Augmentation: dataset’s augmented split + standard image transforms  
+- Early stopping: triggered automatically when validation stopped improving
+
+### Training Curves & Early-Stop Rationale
+- Validation accuracy with EfficientNet-B0 rose steadily and plateaued 
+- Early stopping occurred once validation did not improve (or under what condition AutoGluon decided)  
+- Prevented overfitting while still allowing model to reach its best validation performance
+
+### Test Metrics
+On the held-out **original split** (~40 images):
+
+- **Test Accuracy:** 1.0 
+- **Weighted F1:** 1.0  
 
 ---
 ## Error Analysis