Can we improve information freshness with predictions in mobile crowd-learning?