Plant Disease Classification

Model Audit

Machine Learning Performance

Train/validation behavior and test-set performance, aligned with the original Streamlit analysis.

Accuracy and LossConfusion MatrixMetrics Table

Train, Validation and Test Set: Labels Frequencies

Dataset has Healthy, Powdery, and Rust leaves split into Train (70%), Validation (10%), and Test (20%).

Labels distribution — Label frequencies across train, validation, and test splits

Model History

Accuracy and loss trends suggest a stable fit, with train and validation curves following similar patterns.

Model training accuracy — Training vs validation accuracy trend

Model training losses — Training vs validation loss trend

Generalised Performance on Test Set

Metric	Value
Loss	0.1447
Accuracy	95.14%

Confusion Matrix

Actual \ Predicted	Healthy	Powdery	Rust
Healthy	101	5	1
Powdery	3	96	1
Rust	3	1	98

Model Evaluation Metrics

Metric	Value
accuracy	0.9547
precision	0.9549
recall	0.9547
f1_score	0.9548
specificity	0.9439

Conclusion: the model maintains strong test performance around the 95% target while preserving disease treatment recommendations in the detector flow.