# 6.1 Counterfactual Explanations
A counterfactual explanation of a prediction describes the smallest change to the feature values that changes the prediction to a predefined output.

How do we define a good counterfactual explanation?
1. counterfactual instance produces the predefined prediction as closely as possible
2. counterfactual should be as similar as possible to the instance regarding feature values
3. a counterfactual instance should have feature values that are likely

###### tags: `重點摘要`