With the rapid growth of education data in large-scale assessment, machine learning techniques are crucial to the interdisciplinary development of education and information. Although data mining tools are increasingly used to predict overall student performance, resilient students in the digital world remain unstudied. Our study aims to comprehensively identify key features in the classification of academically resilient students (ARS) and non-academically resilient students (NRS) in digital reading. With a sample of 11,496 disadvantaged students from seven high-performing Asian economies, data drawn from the Programme for International Student Assessment (PISA) 2018 were analyzed through Support Vector Machine (SVM). Results indicated that 20 key features were selected from 105 contextual features at the individual, home, and school levels, which demonstrated a high predictive ability of the model. Personal experience, especially the use of metacognitive strategies in digital reading and reading enjoyment were predominant features. Interestingly, information and communication technology (ICT) resources and usage showed mixed effects on resilient students. This study provides significant implications for cultivating resilient students in online learning environments.