Predicting secondary school students' academic achievement from their elementary school performance and learning behaviours: A longitudinal study based on South Korea's national assessment of educational achievement.

Academic achievement is an important factor strongly related to positive educational experiences that facilitate subsequent learning. Therefore, identifying students who need support at an early stage and promptly providing appropriate intervention play a crucial role in preventing learning deficits. This study examined the longitudinal change in elementary school students' proficiency status to understand the nature of longitudinal transitions. It explored the feasibility of developing early predictive models for identifying students potentially facing academic difficulties based on information obtained from elementary school. Deep neural network and hierarchical generalized linear models were applied to large-scale national assessment panel data collected within the framework of an accountability system in South Korea. The results showed that although most students maintained their initial proficiency status, a substantial number of students made longitudinal transitions from non-proficiency to proficiency or vice-versa as they progressed from elementary to high school. The predictive models had an acceptable level of accuracy in predicting future academic performance, considering the degree of student mobility across proficiency levels over grades. The study concluded that the predictive models can serve as a starting point to identify students who need assistance and provide educators with a means to intervene promptly. [ABSTRACT FROM AUTHOR]