Counterfactual Explanations for XAI Models
Pradeepta Mishra · Apress eBooks · 2021
This chapter explains the use of the What-If Tool (WIT) to explain counterfactual definitions in AI models, such as machine learning-based regression models, classification models, and multi-class classification models. As a data scientist, you don’t just develop a machine learning model; you make sure that your model is not biased and that it is fair about the decisions it makes for new observations that it predicts for the future. It is very important to probe the decisions and verify the algorithmic fairness. Google developed the What-If Tool to address the model fairness issue in machine learning models. You will look at the implementation of the WIT in three ML models: ML for regression-based tasks, ML for binomial classification models, and ML for multi-nominal models.