This paper proposes a new technique for analysing the behaviour of students on an online course. This work considers a range of social learning behaviours supported in our recently designed and implemented collaborative learning system which supports students giving and receiving feedback on each other's developing work and practice. The course was delivered to several thousand students on Coursera during which students were directed onto our social learning environment to take part in group work and assessment activities. This work introduces a swarm intelligence technique, Stochastic Diffusion Search (SDS), and shows how it can be adapted and applied to our data in order to perform classification tasks. The novelty of the approach is not only in using this technique, but also applying it to data linked to "social behaviour" (i.e. how students interact with each other) which differentiates the work apart from many clickstream analysis studies. This paper investigates what combined activity is the best predictor of success or failure in the course. The aim is to argue that the results obtained using the proposed approach indicate the promising potential of predicting students performance through applying swarm intelligence technique to social behaviours. This work has a number of potential benefits including designing better social learning systems, designing more effective social learning and assessment exercises, and encouraging disengaged students. In addition, this work is an important step in addressing our long term goal of evidencing how critical student learning takes place as they give and receive feedback to and from each other on work in progress. [For the full proceedings, see ED592609.]