Extracting Features of Solving Behavior in Handwriting on Tablets

Abstract

This paper proposes a method to estimate the learner's solving behavior during the exercises from the handwriting situation on the tablet. The proposed method extracts the handwriting characteristics in the solving behavior of each learner. It is important to give appropriate guidance according to the understanding of the learner. It is difficult to grasp the understanding status of learners in cases where teachers cannot observe learners directly as in distance learning. Teachers need an online system that estimates the solving behavior of individual learners to notify it of them. In the proposed method, the solving behavior is estimated from logs of both the handwriting and actions seeking supports using the moving window for 30 seconds. The method constructs a random forest model to estimate the solving behavior. Writing characteristics in each solving state are extracted to examine the values of the important variables in a random forest model. Handwriting data were collected from learners in an experiment. The learner's solving behavior was estimated with high accuracy. The constructed model precisely classifies not only states in which the hands stay still from ones in which the hands are moving smoothly but also states of answering correctly from ones of answering incorrectly. Since learners repeating to answer incorrectly are detected on the fly, teachers can take immediate measures for learners who need assistance during exercises.