Towards consistency of rule-based explainer and black box model -- fusion of rule induction and XAI-based feature importance