Resilient students attain high levels of academic achievement despite the presence of chronic socioeconomic disadvantage. Identifying factors that promote resilience in the domain of science is crucial to making equitable and high-quality science education accessible for all students. Rooted in the opportunity-propensity framework, this study examines the relative importance of various opportunity, propensity, and antecedent factors in understanding academic resilience. The data came from 3,377 Grade 8 students in Hong Kong. Among them, 844 students who are in the bottom 25% of SES were selected. Machine learning analyses indicated that ten variables best predicted academic resilience. These variables, in order of predictive power, were: confidence in science, home resources, liking for learning science, valuing of science, instructional clarity, instructional time, content exposure to biology topics, sense of school belonging, school emphasis on academic success, and content exposure to physical science. Mean-level comparisons corroborated the machine learning findings showing that resilient students scored higher on these variables than non-resilient students. This study demonstrates the complexity of academic resilience in science by showing the relative importance of multiple predictors. The findings of the current study could provide policymakers and practitioners in finding the most promising intervention targets to promote academic resilience.