Ad hoc networks require accurate prediction of node mobility in order to be designed and implemented effectively. It is proposed that in this study, a filter-based computing approach is used to predict the movements of neighboring nodes based on their spatial and temporal properties. Reinforcement learning techniques are used to enhance the prediction model's accuracy. Moreover, the greeting message threshold is used to determine which node-finding strategy is the best. A performance evaluation of the proposed welcome message broadcasting algorithm utilizing HP-AODV and ROMSG protocols highlights the cost-effectiveness of this algorithm, which reduces the discovery of neighbor nodes and improves the quality of wireless networks based on ad hoc communication. Despite the complexity of implementing mobility prediction models in the network layer, it has been shown that application-level integration can improve the efficiency of the routing protocol. In this study, we develop a robust framework that can reliably predict the location of wireless devices. The framework is significantly better than the traditional Markov models in various aspects.