Conventional TCP variants do not perform optimally in wireless networks because they assume that any packet loss is primarily due to congestion. As a result, they reduce the transmission rate whenever packet loss occurs, even when it is caused by transmission issues or route failures instead of congestion. These TCP variants are static rule-based and do not have capability to fine-tune their decision making policies dynamically. Over the years, wireless networks are being deployed rapidly in real-world applications. At the same time, networks are becoming more dynamic and diverse in terms of infrastructure and traffic. Researchers have identified necessity for developing more adaptive and generalize TCP to enhance performance under dynamic conditions of networks to detect and handle losses other than congestion losses. Machine Learning based solutions are proposed for TCP protocol for intelligent decision making. In general, ML based TCP variants remain effective but difficult to deploy for resource-constrained wireless networks due to intense computation requirements. PSO (Particle Swarm Optimization) is a promising alternative to ML to adjust TCP parameters such as Cwnd-Congestion Window (transmission rate) dynamically by simulating the social behavior of swarming particles. The algorithm’s global search capabilities explore wide range of configurations to converge on optimal solution without need for extensive data for training. This research work proposes AETCP-PSO (Adaptive Enhancement of TCP based on PSO) to set transmission rate dynamically and optimally. The solution is implemented in Network Simulator 3.28 and has been evaluated with large number of wireless network topologies that are different in terms of connections and traffic patterns. Performance evaluation is performed with end to end transport layer measurements which are average throughput and average packet delivery ratio. A significant performance improvement has been observed for AETCP-PSO over TCP-Westwood+.