Towards Handling Uncertainty in Prognostic Scenarios: Advanced Learning from the Past