Controlling of network data traffic is a big challenge in networks that are wireless. For the management of congestion control, several TCP-based algorithms are proposed in conventional and intelligent ways. When it comes to dealing with a situation that is both complex and critical, there are now intelligent approaches, infrastructure, and algorithms available. By doing so, we will be able to control the congestion through the application of techniques based on machine learning and artificial intelligence. Protocol-based algorithms and different variants are available to improve the congestion but with increasing demand for wireless applications, there is lots of requirement to improve the congestion. With available self-trained models in machine learning congestion can be precisely controlled and improved for the various cases and conditions. This paper proposes a near-distance formula-based deep reinforcement learning algorithm for congestion control. The near-distance formula established the relationship between one host and another host to train network traffic online. The processing of the algorithm reduces the training error of traffic data and improves the rate of latency in the network. An algorithm that was proposed was put through simulation using MATLAB, with the Ad-hoc On-demand Distance Vector (AODV) routing protocol being utilized. Additionally, the algorithm was evaluated using a benchmark dataset. There are a number of different factors that are evaluated for different quantities of nodes. These factors include the ratio of packet delivery (PDR), efficiency, throughput, and queue length. The efficacy of the suggested methodology in comparison to existing operational techniques such as Random Early Detection (RED), Constrained Local Model (CLM), DR-LCC, TCP-Drinc, and Hole Repair Algorithm (HORA). The analysis of the results reveals that the algorithm under consideration is very efficient in terms of throughput and utilization of network resources.