Training vs Testing Data in Machine Learning
Training data is the set of examples the machine learning model learns from. The model studies patterns in the training data during the training process.
About This Resource
Learn why AI models are evaluated on data they have never seen before — and why testing on training data is cheating.
ML model evaluation resource suitable for Class 7–10 students.
What Students Will Learn
- Explain why data is split into training and testing sets
- Define overfitting and explain why it is a problem
- Apply the 80/20 train-test split rule
- Understand model evaluation using a test set
Resources Available
Questions & Answers
What is training data?
Training data is the set of examples the machine learning model learns from. The model studies patterns in the training data during the training process.
What is testing data?
Testing data is a separate set of examples the model has never seen. After training, the model is evaluated on the test set to measure how well it performs on new, unseen data.
Why can't we test on training data?
Testing on training data is like giving students an exam using the exact questions from their textbook — they might get 100% by memorising, but that doesn't prove they truly understand. Similarly, an AI that tests on its training data may appear perfect but fail on real-world data.
What is overfitting?
Overfitting happens when a model learns the training data too perfectly — including the noise and errors — and performs poorly on new data. A good model generalises well: it performs almost as well on test data as it does on training data.
Related Resources
Browse resources by class: