Automated Skin Disease Classifier

Overview

A skin condition classification system that identifies dermatological conditions from images, trained on the Skin Cancer 9 Classes (ISIC) dataset. A ResNet50 backbone clearly outperformed a baseline CNN, reaching around 60% accuracy across 9 classes.

What is included

• A Streamlit web app for image upload and prediction
• Modular training and inference scripts (data.py, model.py, train.py, prediction.py)
• Pretrained ResNet model weights

The lesson that carries into business work

60% across nine classes is a real improvement over the baseline, and it is still not good enough for decisions that matter. The project was a hard lesson in reading results honestly: class imbalance, the cost of each kind of mistake, and why one accuracy number can hide what a model gets wrong. I apply the same rule to every business model: know what an error costs before you put a model in front of a decision.