Predicting learner answers correctness through eye movements with random forest

Abstract

The aim of this research is to predict learners' achievement by using a data mining technique: Random Forest (RF). For this purpose, learners eye movements were recorded by an eye-tracker and their answers to questions were collected via an online assessment tool. Online tests were administered to the students and computer interface was divided into two equal parts, which includes web browser and image processing software. Questions were asked through the browser and participants pencil usage (mouse click counts) was recorded by graphic tablet via the software. Results showed that eye metrics and mouse click counts can be used to predict the answer correctness. While mouse click counts were found to be an important factor for predicting answers in questions that require quantitative operations, fixation count and visit duration metrics are found to be important in questions which include visual elements like graphics. Total fixation duration, number of mouse clicks, fixation count and visit duration were found being the most important eye metrics that predict answers in reasoning questions. Results also showed that changing the presentation modality of a question causes changes in relative importance of each eye metric. © 2014 Springer International Publishing Switzerland.