Stage 5 — Evaluation

Your model works beautifully on the data you trained it with. That proves nothing. Of course it does — it has seen those answers. The real question is what it does with a photo it has never met.

1What evaluation is

Evaluation

In evaluation you test the finished model on newly fetched data — data it has never seen before — and use the results to judge the model and improve it.

The six stages, as a cycle

Tap a stage to stand on it, or press play and watch the work go round. Drag sideways to turn.

5. Evaluation — Test the model on newly fetched data it has never seen, and judge how well it does.

It answers: Does it actually work on data it has not met?

The exam you have already seen

If your teacher gives you the exact paper you practised, your full marks mean nothing about whether you understand the subject. A fair exam has questions you have not seen. Evaluation is the fair exam.

2Why new data, and only new data

A model can pass by memorising instead of learning. It can remember that this exact photo was labelled “diseased” without ever learning what a diseased leaf looks like. Testing on the same data cannot tell the two apart. Testing on new data can, immediately.

Never test on your training data

This is the single most common mistake in an AI project, and the board asks about it. Data used to train the model must be kept separate from data used to test it.

3And then you go round again

If evaluation says the model is not good enough, you do not carry on to deployment. You go back — collect more or better data, or build a different model — and come round the cycle again. This is exactly why the AI Project Cycle is drawn as a circle.

Unit 3 goes deeper

How to measure a model — train-test split, accuracy, error, the confusion matrix, precision, recall and F1 score — is the whole of Unit 3. Here you only need to know what the evaluation stage is and why it uses new data.

4From the board papers

Which stage of the AI Project Cycle involves testing the model on newly fetched data?
(A) Data Exploration (B) Modelling (C) Evaluation (D) DeploymentCBSE board exam 2026

1 mark

▶Show the answer

(C) Evaluation. “Newly fetched data” is the phrase to watch for — it is the definition of evaluation.

Statement 1: The testing data set is a collection of examples given to the model to analyse and learn.
Statement 2: The training data set is used to test the accuracy of the model.CBSE sample paper 2025–26

1 mark

▶Show the answer

Both statements are incorrect — they have been swapped. The training set is what the model learns from; the testing set is what its accuracy is measured on.

5Check yourself

Quick Check

A model scores 100% on the data it was trained with. What does that tell you?

Quick Check

Evaluation says the model is not good enough. What happens next?

← Previous

🧠 Stage 4 — Modelling

Stage 6 — Deployment 🚀