What is Feature Binarization? When to use feature binarization?

This refers to a special case of discretization in which a continuous variable is transformed into a categorical representation that only consists of two bins. Thus, a dummy variable is essentially created, where one bin represents the “on” level and the other the “off” or reference level. This approach works well when there is a near dichotomous threshold that separates observations in relation to the target and also adds fewer dimensions to the feature space compared to if more bins are created.