In this article we propose a proactive scheme that predicts the average queue size, using Forward Linear prediction scheme Recursive Least Square(RLS) and performs Flow Control by the adjustment of TCP window dynamically as per the predicted Queue Size. It is an approach to Congestion Control to improve the network level and user level performance of Voice data over dynamically varying Mobile Ad-hoc Network (MANET). The uncertainty associated with Queue Size estimation and its pitfalls makes this approach play a nominal role in controlling Congestion both proactively and dynamically. The simulation results show that the proposed approach has better performance over the existing schemes for QoS parameters such as Throughput, Packet Dropping Probability, Packet Delivery Ratio, Normalized Routing Overhead and Jitter.