PREDICTION OF STUDENTS’ SCIENCE ACHIEVEMENT: AN APPLICATION OF MULTIVARIATE ADAPTIVE REGRESSION SPLINES AND REGRESSION TREES

Abstract

Turkey is ranked at the 54th out of 72 countries in terms of science achievement in the Programme for International Student Assessment (PISA) survey conducted in 2015, which is a very big disappointment for that country. The aim of this research was to determine factors affecting Turkish students’ science achievements in order to identify the improvement areas using PISA 2015 dataset. To achieve this aim, Multivariate Adaptive Regression Splines (MARS) and Classification and Regression Trees (CART) approaches were used and these approaches were compared in terms of model accuracy statistics. Since Singapore was the top performer country in terms of science achievement in PISA 2015 survey, the analysis results of Turkey and Singapore were compared to each other to understand the differences. The results showed that MARS outperforms the CART in terms of measuring the prediction of students’ science achievement. Furthermore, the most important factors affecting science achievements were environmental optimism, home possessions and science learning time (minutes per week) for Turkey, while the index of economic, social and cultural status, environmental awareness and enjoyment of science for Singapore. Keywords: higher education, machine learning algorithms, PISA, science achievement.