Using machine learning, this research aimed to examine the crucial factors that predict the reading performance of fourth-grade students from Türkiye who participated in PIRLS 2021. When trained with the data of 3589 fourth-grade students and their 405 independent variables, the support vector machine (SVM) algorithm properly distinguished between high- and low-performing students based on 16 key contextual factors at the school, teacher, and family levels. The main factors were at the school level and were related to placing a major emphasis on instruction and the ability of students to borrow books. The teacher-level factors were the assessment strategy, helping students develop reading comprehension skills or strategies, and motivation. The only family-level factor was the parental commitment to ensure that students are ready to learn. Compared to the results of the whole PIRLS 2021 data, the findings of this research revealed a big difference in the key factors predicting the reading performance of fourth graders from Türkiye. Possible reasons were discussed, and new educational policies, interventions, and research practices were suggested. At the policy level, an approach that systemically addresses school, teacher, and family factors may yield more meaningful improvements in reading performance. In terms of interventions, the findings suggest a focus on interactive teaching and assessment strategies that involve students actively interacting with text. As for research practices, this study highlighted the potential of machine learning as a valuable tool to understand the complex, multi-dimensional nature of student performance.
Read full abstract