About the Model

A simple explanation of how DermAI works — in plain language, no complicated jargon.

What is DermAI?

DermAI is an AI tool that looks at a photo of skin and tries to guess what skin condition it might show. It can recognize 23 common skin diseases, like acne, eczema, psoriasis, fungal infections, and even some serious ones like melanoma.

It was built as a student project to show how artificial intelligence can help with healthcare — but it is not a doctor. It’s just a helpful tool to give you an idea, nothing more.

The Dataset (The Photos It Learned From)

The AI learned from a public collection called DermNet, which has about 19,000 real skin photos submitted by people and doctors. These photos are divided into 23 different skin conditions.

Some conditions have many photos (like psoriasis or nail fungus), while others have fewer. This makes it harder for the AI to learn the rare ones, but we used special tricks to help it treat all conditions fairly.

How the AI Model Works

The brain of DermAI is a type of AI called a Convolutional Neural Network (CNN). Think of it like a super-smart photo analyzer that looks for patterns, colors, textures, and shapes in skin images.

• We used a ready-made model called EfficientNetB0 — one of the best and fastest CNNs available.
• It was first trained on millions of everyday photos (like cats, cars, etc.) to learn basic image recognition.
• Then we taught it skin diseases using the DermNet photos.
• We made small changes to the photos during training (flipping, zooming a little) so it learns to handle real-world variations.

How Accurate Is It?

On test photos it has never seen before:

• Top-1 Accuracy: About 41% — it picks the correct disease as the #1 guess 41% of the time.
• Top-5 Accuracy: About 75% — the correct disease is usually in the top 5 suggestions.

This is decent for a challenging dataset with 23 different conditions, but it’s not perfect. Some diseases look very similar (like different types of rashes), which confuses the AI.