Intermediate level · Module 5: Imbalanced and Messy Real-World Data · Lesson 1 of 3
Class Imbalance
Build useful models when the event of interest is rare.
In this lesson you will
- Apply class weights and resampling correctly
- Evaluate imbalanced models with suitable metrics
Key concepts
imbalanceclass_weight='balanced'oversampling and undersamplingprecision–recall evaluation
The lesson
When only 3% of agent-weeks have a shortage, a model can be 97% accurate by always predicting 'no shortage'. Use recall, precision, F1 or average precision instead of accuracy.
Many models accept class_weight='balanced', which makes errors on the rare class count more. Resampling — duplicating rare cases or removing common ones — is an alternative, but must be done only on training folds.
Often the simplest effective step is adjusting the decision threshold, as chosen from costs in the previous module.
Step by step
- Measure the baseline's accuracy and recall.
- Train with class_weight='balanced'.
- Try oversampling inside the pipeline's training folds.
- Compare average precision across approaches.
Practice activity
Choose the best approach to imbalance for the cash-shortage model and justify it.
Expected result: A comparison table and a justified choice using imbalance-appropriate metrics.
Check your understanding
-
Why is accuracy misleading for rare events?
Show answer
Always predicting the common class gives high accuracy but finds no events.
-
Where must resampling happen?
Show answer
Only on the training part of each fold, never before splitting.
Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.