Robust Counterfactual Explanations under Model Multiplicity Using Multi-Objective Optimization

Artificial intelligence(AI), including machine learning, is used in many domains. However, although many machine-learning methods have high prediction accuracy, they are often considered'black boxes' because the processes involved are unclear owing to their complex combination of nonlinearities and interactions. Explainable AI or interpretable machine learning has become an important issue in addressing these problems [1, 7, 18]. Several such methods are available. One such method is white-box machine learning. There are also methods for ensuring the interpretability of black-box machine learning. They examine which variables are important in the overall data and which variables are important in individual data. Among these methods, one is called the counterfactual explanation (CE) [10, 14, 27]. CEs are outputs that indicate that, for a trained supervised machine-learning model, the minimum changes to the original data (explanatory variables) are needed to achieve a particular desired predictive outcome.