Machine Learning · Lesson 7
Overfitting, Underfitting & Model Generalisation
Learn to diagnose a model that is too simple, too complex, or unable to perform reliably on unseen data. Practise with business scenarios and interactive checks.
The goal: generalisation
A model should work on data it has never seen.
Machine learning is useful when it learns a pattern that continues to hold beyond the examples used during training. This ability is called generalisation. A model that memorises training examples may look excellent during training but fail on new customers, new months, or new locations.
Business meaning
A retail model should forecast next month's demand—not merely reproduce last month's sales.
Generalisation checklist
Try it
A demand model scores 99% on training data but only 62% on a fresh test set. What is the strongest conclusion?
Quick knowledge check
What does generalisation mean?