Identifying students facing difficulties and providing them with timely support is one of the educator's key responsibilities. Yet, this task is becoming increasingly challenging as the complexity of physical learning spaces grows, along with the emergence of novel educational technologies and classroom designs. There has been substantial research and development work focused on identifying student social behaviours in digital platforms (eg, the learning management system) as predictors of academic progression. However, little work has investigated such relationships in physical learning spaces. This study explores the potential of using wearable trackers for the early detection of low-progress students based on their social and spatial (socio-spatial) behaviours at the school. Positioning data from 98 primary school students and six teachers were automatically captured over a period of eight weeks. Fourteen socio-spatial behavioural features were extracted and processed using a set of machine learning classifiers to model students' learning progression. Results illustrate the potential of prospectively identifying low-progress students from these features and the importance of adapting classroom learning analytics to differences in pedagogical designs.