Automated Scoring of an Interactive Geometry Item: A Proof-of-Concept

Abstract

An online interactive geometry item was developed to explore students’ abilities to create prototypical and “tilted” rectangles out of line segments. The item was administered to 1,002 students. The responses to the item were hand-coded as correct, incorrect, or incorrect with possible evidence of a misconception. A variation of the nearest neighbor algorithm was used to automatically predict one of these categories for each of the student responses. The predicted category was compared to the hand-coded category. The algorithm accurately predicted the category for 94.6% of responses.