Classifying Students' Performance in Mathematics in a Multicultural Primary School Using Machine Learning Algorithms

The last few years have witnessed a global change in the education system that was exacerbated with Covid-19 pandemic. The increasing use of online learning resources has introduced a challenge in deliverying quality education, especially in multicultural societies. In this respect, the first step to delivering high-quality education involves optimising educational quality by identifying facilitators and barriers to it. This research proposed the use of machine learning (ML) algorithms to understand the influence of key characteristics on the performance of level 6 students at a primary school in Mathematics. Three ML algorithms were applied to 12 characteristics related to students’ performance over three semesters being autumn, Spring and Summer. The ML algorithms were correlation in variable space method (CM), principal component analysis (PCA) and self-organizing maps (SOM). The aforementioned 12 characteristics included: attendance, behavior, engagement, nationality, previous school, age, weekly homework, daily in-class exercise, previous report, gender, learning disability, benchmark testing and end of block assessment. The results showed that the influence of characteristics was related to the type of assessment/lessons undertaken by students. In all cases, five characteristics played a key role and included attendance, weekly homework, daily in-class test, previous report and benchmark testing. The extent to which degree these five characteristics influenced performance varied between lessons depending on the type of task undertaken. Overall, the performance of students was consistently similar across the different semesters. Future work involves exploring the prediction of student performance based on the proposed 12 characteristics.