Knowledge Maxxing← Back to topic library
ML MaxxingBeginner60 sec
Generalisation
Generalisation is how well a model's learned patterns apply to new data it has never seen, rather than only the data it was trained on.
Prerequisite: Training set
What matters
The entire goal of training a model is generalisation; a model that only performs well on its training data provides no value on new, real-world inputs.
Example
A model that scores 99% on training data but only 60% on new data has poor generalisation, despite its training-set score looking excellent.
Common misconception
High training-set accuracy is not the same as good generalisation; the two can move in opposite directions as a model overfits.
Recall question
Why is training-set accuracy alone not evidence of good generalisation?
Keywords and related concepts
generalisationunseen datamodel qualityoverfitting