SHAP-based Explanations are Sensitive to Feature Representation