The efficiency of routing in ad-hoc networks depends on the node traffic. One of the methods of improving the network efficiency in MANETs is by predicting the network traffic. It is very important to predict the characteristics of the future network traffic from the previous parameters. In this paper, we propose TPAM-a a multipath routing based traffic prediction algorithm that uses RNN architecture. TPAM consists of two modules, which includes a multipath routing algorithm and a network congestion discovery using RNN. It is clear from the simulation results that RNN architecture provides promising results in predicting the network traffic under varying conditions. Further, the algorithm has improved efficiency in routing by using the multipath selection method. Finally, the proposed algorithm has a less end-to-end delay, lower overhead and a high success ratio.