Improving the Explainability of Multi-criteria Decision-Making Using Neutrosophic Logic

In recent years, explainability in machine learning has been an area of significant interest. As previous research came to show, opaque models, lack the level of interpretability required to provide meaningful explanations. Therefore, further research is needed on interpretable models. However, focusing on interpretability comes with its challenges, such as the perceived performance tax. In this paper, neutrosophic logic and multi-criteria-decision making are combined to construct a data-driven explainable model for classification. Consecutively, the proposed model is compared with several state-of-the-art classifiers using the Breast Cancer Wisconsin (Original) benchmark dataset from the UCI repository [1]. The model achieved a specificity of 79% and a sensitivity of 99%. Moreover, the results demonstrate that the model’s transparent nature allows for generating a meaningful textual and graphical explanation of the classification decision. However, there is an observed trade-off between enhanced explainability and predictive accuracy.