Automated Clock Drawing Test through Machine Learning and Geometric Analysis

In this paper, we discuss the challenges of sketch recognition accuracy and automation of the Clock Drawing Test (CDT). Sketch recognition in the context of the CDT is a complex problem due to the lack of knowledge of the preference bias among the sketches drawn by neuro-atypical patients. However, machine learning provides a viable solution to detect measurable patterns among sketches drawn in the CDT. The paper posits that these sketches are constrained, and the majority of the recognition work can be completed by doing handwritten digit recognition. All other properties to be measured in the CDT are primarily geometric in nature. The paper explores various machine learning classification techniques for handwritten digit recognition. Finally, based on analysis, we present the best one in context of CDT.