Wireless Network is one of the Internet-of-Things (IoT) prototypes that come up with monitoring services, therefore, influencing the life of human beings. To ensure efficiency and robustness, Quality-of-Service (QoS) is of the predominant point at issue. Congestion in wireless networks will moreover minimize the anticipated QoS of the related applications. Motivated by this, a novel method called, Ornstein–Uhlenbeck Transition and Cache Obliviousness Neural Adaptive (OUT-CONA) to improve congestion control of wireless mesh networks is presented. Adaptive actor-critic deep reinforcement learning scheme on Ornstein–Uhlenbeck State Transition scheduling model to address handovers during data transmission for IoT-enabled Wireless Networks is first designed. Here, by employing the Ornstein–Uhlenbeck state transition scheduling model, both the advantages of the Gauss and Markov Processes are exploited, therefore reducing the energy consumption involved while performing the transition. Next, in the OUT-CONA method, LSTM is imposed for learning the current state representation. The LSTM with the current state representation achieves the objective of controlling congestion with cache obliviousness. The Cache Obliviousness-based Congestion method is utilized for congestion control with obliviousness caching using coherent shielding among organized as well as disorganized data. Furthermore, the performance of the OUT-CONA method is evaluated and compares the results with the performances of conventional techniques, adaptive aggregation as well as hybrid deep learning. The evaluation of the OUT-CONA congestion control method attains better network using lesser misclassification rate, consumption of energy, delay as well as higher goodput using conventional methods in Wireless Mesh Networks.