Model Training & Adaptation
Class imbalance
Having many more learning examples of some groups than others.
Example
A fraud-checking collection contains far more ordinary payments than dishonest ones.
Why people use it
It warns that rare cases can be overshadowed by common ones during learning or testing.
What you'll hear
“Almost every example is in the same category.”
What this means for you
Check results for each group and the practical cost of mistakes.
Can you control it?
No
No direct control. This describes a wider issue, concept or result rather than something you can simply switch on or off in a tool.
Common questions
- Can high accuracy hide poor minority-group performance?
- Yes. Predicting the common group can look accurate while missing rare cases.
- Can changing the training mix help?
- Sometimes. Giving rare cases more attention can help, but it can also change the kinds of mistakes made.
- Does balancing the examples make every case equally important?
- No. The consequences of different mistakes remain a separate question.