Imbalanced data commonly exist in real world, especially in anomaly-detection tasks. Handling imbalanced data is important to the tasks, otherwise the predictions are biased towards the majority class. RandomUnderSampler, ClusterCentroids, CondensedNearestNeighbour, and etc. are useful undersampling tools to remove data for majority classes.
The data for the "Poor" and "Good" classes are much less than the "Standard" class:
The predictions are biased towards the majority class:
Undersampling with RandomUnderSampler:
The predictions get more balanced:
