Confidence over enjoyment? A cross-national explainable machine learning analysis of affective predictors of early science achievement

Background A growing body of research has examined students’ science achievement using large-scale international assessments; relatively few studies have focused on affective–motivational factors in early grades within a cross-national and machine-learning framework. Purpose This study aimed to investigate the extent to which affective–motivational factors predict Grade 4 students’ science achievement and to examine whether these relationships differ between the United Arab Emirates (UAE) and Singapore (SGP). Sample The study utilized TIMSS, 2023 Grade 4 data from the UAE and SGP, comprising a weighted sample of 34,842 students from the UAE and 6,530 students from Singapore. Design and methods Four affective–motivational constructs, students’ confidence in science, digital self-efficacy, enjoyment of learning science, and value of environmental preservation, were examined as predictors of science achievement. Five plausible values of science achievement were modeled separately following TIMSS methodological guidelines. A comparative machine-learning framework was applied using linear regression, lasso regression, decision trees, random forests, and neural networks. SHAP was used to describe how each predictor contributed to the fitted model’s predictions and how these contributions differed between countries. Result Random forest models demonstrated the strongest predictive performance, although overall explained variance was modest (R 2 ≈ 0.14–0.15). Students’ confidence in science emerged as the most influential predictor in both countries, followed by digital self-efficacy, enjoyment of science, and environmental values. SHAP descriptions of the fitted model suggested that the contribution of enjoyment of science differed between countries, being larger at high levels of enjoyment in the UAE; given the modest predictive accuracy of the models, these patterns are interpreted as exploratory. Conclusion The findings suggest that affective–motivational factors provide meaningful but limited predictive information about Grade 4 science achievement. Differences in predictive patterns between the UAE and Singapore indicate that the relationships were not identical across the two national samples; however, the study does not identify the contextual mechanisms underlying these differences.