Motion capturing is the de facto standard for objective measurement of motion patterns in clinical applications. However, the arising data is hard to interpret and visualize in an intuitive manner. To aid physicians in diagnostic decisions, we present a pipeline that predicts diseases based on motion capture gait sequences and visualizes the most important features that led to classification decision in an intuitive manner. To account for arbitrary motion capture systems, we transfer the motion capture data into a common reference frame, in which we map the gait sequence into a unified joint-based space. This representation can not only improve the neural network based classification but also enables the projection of interpretability metrics from the classifier back onto the reference shape model, which ultimately is visualized in a human interpretable way. The proposed pipeline allows clinicians to comprehend decisions made by our classifier even if they are based on subtle changes in movement patterns. We evaluate our approach on a Parkinson’s disease classification task demonstrating its applicability and interpretability. The unified joint-based representation is very general and can also be applied for e.g. hands or be deployed in settings where the 3D motion is retrieved via existing 3D-from-2D methods.