Spatial visualization skills are essential and fundamental to studying STEM subjects, and sketching is an effective way to practice those skills. One significant challenge of supporting practice using sketching questions is the vast number of possible mistakes, making it time-consuming for instructors to provide customized and actionable feedback to students. The same challenge persists for computer programs as well. This paper introduces a clustering model designed to categorize sketching answers based on the severity and characteristics of their mistakes. The model is designed to be used by a computer-based training platform to provide customized, actionable formative feedback to students in real-time. The promising results also suggest a new and comprehensive set of evaluation criteria to assess a student's performance on sketching questions. As a broader contribution, our work is a proof-of-concept for a modeling approach to automatically evaluate and provide formative feedback on complex free-hand sketches using abstract features that may be generalized to a variety of disciplines that involve the creation of technical drawings. [For the full proceedings, see ED607784.]