Using machine learning approaches to predict Taiwanese eighth graders' computational thinking performance in ICILS 2023 study

This study employs machine learning approaches to examine how socio-demographic, student-related, and school-related variables predict the computational thinking (CT) performance of 5211 Taiwanese eighth graders in the ICILS 2023 study (Fraillon, 2024). It further aims to identify the key predictors of Taiwanese students' CT scores in this international evaluation project. The study used seven trained models: Multinomial Logistic Regression, Random Forest, AdaBoost, XGBoost, LightGBM, Gradient Boosting classifier, and Stacking Ensemble to identify and rank the variables that affect CT scores. The CT performance score was used as a binary variable with two classes: below and above average score. Findings showed that XGBoost and Stacking Ensemble performed best when classifying below and average CT scores respectively in terms of precision, recall and F1 score. In addition, among the variables, student-related variables had the highest impact on students' CT skills followed by school-related and socio-demographic. Among student-related variables, CT disposition was the most significant variable followed by ICT self-efficacy and academic multitasking. Further, among school-related factor, learning special applications in class had significant impact followed by a low impact of socio-demographic variables such as home literacy and parents' education. This study offers practical implications for educators, policymakers, and curriculum designers by underscoring the role of CT disposition and recommending targeted support for enhancing students’ digital self-efficacy. Additionally, the study shows the potential of ML for creating adaptive learning environments and guiding data-informed decisions in educational policy and practice.