The delivery rate in delay tolerant networks is directly impacted by the buffer management algorithm. Recently, by considering the social characteristics of the network node brought a new point of view to the design of such algorithms. This work aims to propose an algorithm that takes into account the selfishness of each node when it has to discard a message from its buffer. A learning machine technique estimates the friendship strength among the nodes. To improve this classification, the

more »

... ed algorithm makes use of a training database from an experiment carried out with users from the MIT campus. By means of stochastic simulation, the validation tasks used the Reality and Cambridge traces for the sake of comparison with similar works. The results showed that the proposed algorithm increased the delivery rate while reducing both the average delivery time and the number of hops to deliver the message to the destiny.