Towards Refined Classifications Driven by SHAP Explanations