Online education offers students increased flexibility to access course materials from virtually any location. While this mode of learning presents several advantages, it also introduces new challenges. The ability to attend classes from any environment means students are not always situated in spaces conducive to focused learning, increasing their susceptibility to distractions. In addition, the online format introduces barriers that make it more difficult for instructors to accurately assess student attentiveness. Therefore, early detection of distractions is essential not only to support students in retaining information more effectively, but also to enable educators, content creators, and employers to implement timely interventions that promote sustained attention and improve learning outcomes. This paper builds on our previous work that identified body motion and physiological signals as indicators of attention in an online setting. In that study, we collected and analyzed aggregated body motion and physiological data from twenty participants who each completed four 30-min sessions—one control and three experimental trials with varying degrees of distractions. In this study, we apply machine learning to raw, unaggregated signal data to gain deeper insights into attention dynamics. The results highlight the feasibility of using bioindicators and accelerometer data in predicting student attention. We demonstrate that machine learning models, especially random forest, gradient boosting, and neural networks, can accurately predict student attention using both physiological and motion-based data, achieving up to 97% accuracy. These findings suggest a promising path toward real-time attention monitoring tools that can enhance learning outcomes in online education.