Predicting examination performance using machine learning approach: A case study of the Grade 5 scholarship examination in Sri Lanka

Abstract

Universal primary school education is a must requirement and one of the criteria that should be fulfilled by the developing countries according to the International development goals which are also recognized as “Eight Millennium Development Goals”. In the context of Sri Lanka, Government is mostly involved in primary education through government-controlled schools. The success of primary education is measured by conducting a scholarship examination. Those who are getting higher results are given opportunities to attend well-facilitated schools for secondary education. Due to that case, there is a massive competition for passing the examination. Limitless pressure for examination provides lots of issues to students. This paper uses data to investigate a model of academic performance as measured by past results of school tests of Grade 4 and Grade 5. 500 students from eight primary schools in the Gampaha district have been selected for collecting data. The Data on the above-mentioned students have been collected by conducting questionnaires to the teachers who incharged the classes. Then the Logistic Regression model and Multiple Linear Regression model have been applied to predict students’ performances at the examination. The model depicts the likelihood of a student passing or failing the grade 5 scholarship examination and predicts the range of results that students will obtain in the examination. The accuracy of predictive models is measured using the results of students who have already faced the Grade 5 examination. Revealing the potential of students at the grade 5 examination is heavily benefited by teachers because they can provide personalized education for talented students and provide opportunities to other students to improve their talents. The initial architecture of the Grade 5 examination results’ predictive model is being discussed in this paper.