Differential Pattern Mining of Students' Handwritten Coursework

A key challenge in educational data mining research is capturing student work in a form suitable for computational analysis. Online learning environments, such as intelligent tutoring systems, have proven to be one effective means for accomplishing this. Here, we investigate a method for capturing students’ ordinary handwritten coursework in digital form. We provided students with Livescribedigital pens which they used to complete all of their homework and exams. These pens work as traditional pens but additionally digitize students’ handwriting into time-stamped pen strokes enabling us to analyze not only the final image, but also the sequence in which it was written. By applying data mining techniques to digital copies of students’ handwritten work, we seek to gain insights into the cognitive processes employed by students in an ordinary work environment. We present a novel transformation of the pen stroke data, which represents each student’s homework solution as a sequence of discrete actions. We apply differential data mining techniques to these sequences to identify those patterns of actions that are more frequently exhibited by either goodor poor-performing students. We compute numerical features from those patterns which we use to predict performance in the course. The resulting model explains up to 34.4% of the variance in students’ final course grade. Furthermore the underlying parameters of the model indicate which patterns best correlate with positive performance. These patterns in turn provide valuable insight into the cognitive processes employed by students, which can be directly used by the instructor to identify and address deficiencies in students’ understanding.