Capturing both the structural and temporal aspects of interactions is crucial for many real world datasets like contact between individuals. Using the link stream formalism to capture the dynamic of the systems, we tackle the issue of activity prediction in link streams, that is to say predicting the number of links occurring during a given period of time and we present a protocol that takes advantage of the temporal and structural information contained in the link stream. Using a supervised learning method, we are able to model the dynamic of our system to improve the prediction. We investigate the behavior of our algorithm and crucial elements affecting the prediction. By introducing different categories of pair of nodes, we are able to improve the quality as well as increase the diversity of our prediction.