On the explainability of quantum neural networks based on variational quantum circuits