Class Imbalance

When there are more sample from one class than the other. Can lead to poor results in Classification and Regression tasks as the rare class hasnt been seen often by the model while training.

Methods to circumvent this for binary classification problems are:

• Oversampling
• Undersampling
• SMOTE
• Moving the Decision Boundary so that rare samples are easier to classify (works for Bayesian Classifier, Naive Bayes Classifier)
• Ensemble techniques

For multiclass problems it is still hard to remove class imbalance artifically without skewing their relation.