Verification of Usefulness of Student Modeling with Real Educational Data using Convex Factorization Machines

Abstract

Extracting useful information generated from educational settings involves the application of data mining, machine learning, and statistics to the large amount of electronic data collected by educational systems. To generate better higher learning outcomes using an intelligent tutoring system, such as an e-learning system, it is necessary to more accurately understand the state of student knowledge. The purpose of student modeling is to estimate the students’ skills from log data, such as examination results, and to predict whether or not a student will be able to solve a problem.In this study, we propose a student performance prediction method using convex factorization machines. Factorization machines offer a combination of the advantages of support vector machines and factorization models such as matrix factorization. The results of conventional methods, which predict student performance using factorization machines, exhibited better results than they have before. However, because factorization machines are not convex optimizations, they acquire local minima, which is a disadvantage. Therefore, we used convex factorization machines in order to improve the performance of student modeling predictions.