A framework consisting of a deep tree model and a long short-term memory network is proposed to predict future traffic congestion. Based on the prediction, the uplink and downlink ratio is adapted to ...