Human-interpretable model explainability on high-dimensional data