In this paper we propose a new artificial neural network model for TCP congestion control based on four parameters 1 loss event rate (p) 2. Round trip time (RTT) 3 retransmission time out (RTO) 4 numbers of packets acknowledged by an arriving ACK (b).we believe that with inclusion of b proposed neural network model will more accurately estimate TCP throughput. In new concept of ACK compression in wireless networks arriving ACK can acknowledge more than one packet and definitely influence the behavior of TCP. After training on 500 samples, a three layer (4-16-1) artificial neural network model has been tested over variety of network scenarios in comparison to equation model and previously proposed neural network model, over proposed model can better associate TCP factors. As this model also implements online learning so it can better adopt to new trends.