Planning sport sessions automatically is becoming a very important aspect of improving an athlete’s fitness. So far, many Artificial Intelligence methods have been proposed for planning sport training sessions. These methods depend largely on test data, where Machine Learning models are built, yet evaluated later. However, one of the biggest concerns of Machine Learning is dealing with data that are not present in the training dataset, but are unavoidable for predicting the further improvements. This also represents a bottleneck in the domain of Sport Training, where algorithms can hardly predict the future training sessions that are compiled with the attributes of features presented in a training dataset. Usually, this results in an under-trained trainee. In this paper we look on this problem, and propose a novel method for synthetic data augmentation applied on the original dataset.