An unmanned aerial vehicle (UAV) network can serve as an aerial relay to periodically receive packets from macro base stations (BSs). Severe packet loss may happen especially when UAVs have bad wireless connections to a BS. In this paper, a data exchange scheme is proposed utilizing unsupervised learning to enable efficient lost packet retrieval through reliable wireless transmissions between UAVs instead of through retransmissions of macro BSs with a longer delay and higher overhead. With the proposed scheme, all UAVs are assigned to multiple clusters and a UAV can only request its lost packets to UAVs in the same cluster. By this way, UAVs in different clusters could carry out their lost packets retrieval processes simultaneously to expedite data exchange. The agglomerative hierarchical clustering, a type of unsupervised learning, is used to conduct clustering, guaranteeing that UAVs clustered together could supply and supplement each other's lost packets. To further enhance data exchange efficiency, a data exchange mechanism is designed, where the priority of UAVs performing data exchange is determined by the number of their lost packets or the number of requested packets that they can provide. The introduced data exchange mechanism can make each request-reply process maximally beneficial to other UAVs' lost packet retrieval in the same cluster. A new random backoff procedure based on the carrier sense multiple access with collision avoidance (CSMA/CA) is designed to support the data exchange mechanism. Finally, simulation studies are conducted to demonstrate the effectiveness and superiority of our proposed data exchange scheme.