Communication is a key component of everyday vehicle-pedestrian-interaction. As pedestrians feel unsafe when not communicating with a driver, this will become a major challenge for the introduction of automated vehicles. Explicit communication using symbols, signs and colors is not recognized intuitively, which is why a learning process needs to take place. The meaningfulness of explicit cues and the learning process for these can be supported using implicit communication. To identify natural, implicit communication cues, we use model-agnostic methods to interpret a machine learning-based algorithm. This algorithm predicts the driver's intention to stop at zebra crossings. We use various interpretations applying feature permutation to identify five most important features for implicit vehicle cues at crosswalks. These are velocity, x/y-accelerations, steering wheel = angle and brake pressure. We then utilize an example-based interpretation, which results in time series for implicit communication at cross walks. Automated vehicles can apply these implicit cues together with explicit light based communication to achieve an intuitive interaction with pedestrians or support the learning process of these newly introduced communication forms.