On the stability, correctness and plausibility of visual explanation methods based on feature importance