It can be relevant for clinicians to have access to a prediction of the time to an event such as a relapse, a cancer occurrence, or the death of a patient. When predicting the time to event based on right-censored data, it is natural to rather consider a restricted time because of tail estimation issues. The prediction task is then equivalent to the estimation of the restricted mean survival time (RMST). To that aim, we propose a new flexible and easy to use regression model based on pseudo-observations and super learning. To prove the theoretical validity of this method, we present a new definition of the pseudo-observations.