Fuzzy multi-criteria decision-making: example of an explainable classification framework

Explanation, or system interpretability, has always been important in applications where critical decisions need to be made, for example in the justice system or biomedical applications. In artificial intelligence and machine learning, there is an ever increasing need for system interpretability. This paper investigates a Fuzzy Multi-Criteria Decision-Making (MCDM) model as the basis for an interpretable framework for explainable classification. The proposed framework includes a Fuzzy Inference System paired with a modified MCDM-based model for data-driven classification. The modular nature of MCDM allows for the development of a model-based layer capable of generating factual and counterfactual explanations. Results on a ‘Titanic’ survivors’ dataset classification, which illustrates a minimal trade-off in predictive performance while gaining textual and graphical explanation, autonomously provided by the proposed model-based MCDM framework.