Medical Image Classification — Model Comparison

Comparison of ML classifiers for medical image classification tasks. Covers the full pipeline: data loading, feature extraction, training, evaluation, and a prediction API.

What this covers

• Statistical feature extraction from image arrays
• Three classifier architectures: Random Forest, Gradient Boosting, SVM (RBF)
• Accuracy and F1 comparison with bar chart output
• Prediction API function ready for deployment

How to run

pip install numpy scikit-learn matplotlib
python model_comparison.py

Swap in a real dataset

Replace generate_synthetic_data() with any of:

• MedMNIST: pip install medmnist — standardised medical imaging benchmarks
• Chest X-Ray14 (NIH): 112,000 chest X-ray images
• ISIC skin lesion: dermoscopy image classification

Replace ImageFeatureExtractor with a CNN backbone (ResNet50, EfficientNet) from PyTorch/TensorFlow for production-level performance.

Results (synthetic data)

Model	Accuracy	F1 (weighted)
RandomForest	~0.44	~0.43
GradientBoosting	~0.46	~0.46
SVM RBF	~0.45	~0.44

Results on synthetic data are near-random (expected). On real medical datasets, CNN backbones achieve 85-95%+ accuracy.

Skills demonstrated

Python, NumPy, Scikit-learn, Matplotlib, ML model comparison, prediction API design