The current paper proposes a novel end-to-end congestion control suitable for multimedia flows, with the primary aim of guaranteeing a good level of smoothness to the connections that employ it. The suggested strategy, termed Adaptive GAIMD, has as its starting point a blind, window-based scheme named General Additive Increase Multiplicative Decrease (GAIMD); it however enhances it through a measurement-based prediction of network conditions, based on the observation of the interarrival times of the most recently received triple duplicate acknowledgments. Such forecast allows the control to selectively switch to the AIMD policy, i.e., to a TCP-like behavior, when the network drifts towards overload conditions. On the other hand, if the network is sensed lightly loaded, the control behaves as GAIMD. In all cases, Adaptive GAIMD guarantees a smoother packet delivery than native GAIMD and TCP would. The validity of the proposed approach is experimentally proved, performing numerous tests and comparisons between Adaptive GAIMD, GAIMD and TCP-Sack, that took place on a geographical link spanning from Modena, Italy, to Los Angeles, United States.