Abstract
This article explored the influencing factors of digital reading achievement based on the PISA 2018 assessment of students' reading achievement. An integrated Random Effect-Expectation Maximization (RE-EM) regression tree model was the first constructed to address the shortcomings of traditional machine learning methods for nested data estimation and the limitations of traditional linear models in handling complex data. Our study identified the key variables for the feature selection in the integrated RE-EM regression tree model include various aspects of Meta-cognition, as well as the affective element of Joy/Liking for Reading. Notably, this study found that Meta-cognition: Assess Credibility exhibits a ceiling effect on reading achievement, where the marginal effect on reading achievement significantly diminishes at the higher variable values. Additionally, Meta-cognition: Summarizing and Joy/Liking for Reading both demonstrate an approximately S-shaped curve influence on reading achievement. These findings were discussed in critical theoretical and policy implications.