Teacher-Involved Automatic Characteristics Classification in Handwritten Math Answer Process

Abstract

In Japanese junior high schools, while digital logs of students' handwritten math answers offer utilization opportunities for classroom sharing, manually reviewing them is burdensome for teachers, and existing automatic classification methods often fail to meet their pedagogical needs. This study, co-designed with teachers, definedfour pedagogically classification labels, identified their associated handwriting features, and subsequently evaluated classification models. Among the models tested, XGBoost was most effective, notably meeting our success criterion (precision > 0.5) for teacher support on the diagram problem type (0.532), thus demonstrating the practical feasibility of this approach. Our findings highlight the importance of a teacher-involved approach to designing learning analytics systems for practical classroom use.