Enhancing interpretability of rule-based classifiers through feature graphs