Abstract: For the load balancing of ad hoc networks, a routing protocol based on traffic prediction with wavelet neural network (WNNP-LBRP) was proposed, where the number of packets buffered in the interface of the MAC layer was a measure of node traffic. The node traffic and its dynamic threshold were calculated using the wavelet neural network prediction model and then were compared with each other in order to avoid taking a heavy-load node as an intermediate node, which would result in network congestion. Thus, the link could be updated before congestion and the network load could be distributed evenly. Simulation results showed that the WNNP-LBRP outperforms the LBR-AODV and AODV, that is, the packet delivery ratio is increased and the end-to-end delay and routing overhead are reduced.