🎯 Model Accuracy

5th Grade 🤖 Artificial Intelligence

Accuracy is a measure of how often an AI model gives the right answer. For example, if a model was right on 90 out of 100 test pictures, its accuracy is 90 percent.

Free · US Letter · no sign-up
Google Classroom Save to Pinterest Email me

To measure accuracy, you test the model on examples it didn’t see during training and that you already have the right answers for. You count how many times it was right and divide by the number of examples. That way you can compare models and check whether a change you made improved the model or made it worse.

Even a very accurate model makes mistakes. 99 percent accuracy sounds great, but if the model checks a million pictures, it will get about ten thousand of them wrong. It also matters which cases it gets wrong: a medical model that misses an illness makes a more dangerous mistake than a model that names a flower wrong.

A common mistake is thinking that high accuracy on a test guarantees success everywhere. If pictures in the real world are different from the test pictures — different lighting, a different camera or different people — the accuracy can drop. That’s why models keep being checked even after people start using them.

For example

  • A model that correctly recognized 45 out of 50 fruit pictures — 90 percent accuracy.
  • Comparing two spam filters to pick the more accurate one.
  • A model that recognizes daytime pictures well but nighttime pictures less well.
  • A medical system that gets checked especially for how many important cases it misses.

📌 Remember: Accuracy = how often the model is right. Even 99 percent means there are mistakes.

Quick quiz

  1. A model was right on 80 out of 100 tests. What is its accuracy?

  2. Which examples do you measure accuracy on?

  3. A model is 99 percent accurate. Does it never make mistakes?

🔎 Ask Thinkapedia another question