Connected devices such as smartwatches and fitness wristbands are getting more and more popular producing more and more data everyday. Runners use these devices to track all their movements. Using a Long Shot-Term Memory neural network, this paper shows that recurrent neural network can be used to keep in memory implicit features like the tiredness of a runner and predict his speed on a known path. The results show a small loss in the training and evaluation process, however, when plotting the predicting run over the original run, the plots are quite different with few resemblances. The model is trained over a sample of runs and with a simple architecture with less than 200 parameters. Further work is required with the whole dataset available, on a longer period of time for the training phase and a more complex model.