Towards Improving Human Arithmetic Learning using Machine Learning

Abstract

Basic arithmetic is an essential skill that is used in almost all career paths in some way. Ensuring that young children have a solid foundation in simple mathematical concepts is a worldwide goal and new methods to improve arithmetic learning are constantly being developed. Our aim is to utilise machine learning to assist learners with developing their basic mathematics skills by identifying the types of problems a user struggles with and presenting them with targeted questions to improve in these areas. In this paper we focus only on the prediction component: given a set of arithmetic questions and corresponding answers, can we predict which future questions a user will answer incorrectly? The accuracy and suitability of four machine learning models are evaluated using data from computer-generated agents as well as human users. On simulated agents, our models achieve accuracies of around 79% to 96% with decision trees performing the best. On human data, our models achieve accuracies in the range of 63% to 69%, with the decision tree once again outperforming other approaches. We hope that these error predictions models could be incorporated into future E-learning systems targeted at human arithmetic learning.