Symbolic knowledge-extraction (SKE) techniques are becoming of key importance for AI applications since they enable the explanation of opaque black-box predictors, enhancing trust and transparency. Among all the available SKE techniques, the best option for the case at hand should be selected. However, an automatic comparison between different options can be performed only if an adequate metric – such as a scoring function resuming all the interesting features of the extractors – is provided. Regrettably, the literature currently lacks definitions of effective evaluation metrics for symbolic knowledge extractors. This paper proposes the novel FiRe score metric, which comprehensively assesses the quality of an SKE procedure by considering both its predictive performance and the readability of the extracted knowledge. FiRe is compared to another existing scoring metric and a rigorous mathematical formulation is provided along with several practical examples to highlight its effectiveness to the end of being exploited inside automatic hyper-parameter tuning procedures.

This paper proposes CREAM, a new explainable clustering technique based on decision tree induction, providing human-interpretable clusters by performing hypercubic approximations of the input feature space. CREAM may also be applied to data sets describing classification and regression tasks, given that the algorithm discriminates amongst input and output features. We also present OrCHiD, an automated tuning procedure to select the optimum CREAM parameter. Experiments demonstrating the effectiveness of CREAM in clustering, classification, and regression tasks are reported here, in comparison with other state-of-the-art techniques used as benchmarks.