Different high speed Transport layer protocols have been designed and proposed in the literature to improve the performance of standard TCP on high BDP links. They are mainly different in their increase and decrease formulas of their respective congestion control algorithm. Most of these high speed protocols consider every packet drop in the network as an indication of congestion and they immediately reduce their congestion window size. Such an approach will usually result in under utilization of available bandwidth in case of noisy channel conditions. We take CUBIC as a test case and have compared its performance in case of normal and noisy channel conditions. The throughput of CUBIC was drastically degraded from 50Mbps to 0.5Mbps when we introduced a random packet drops with 0.001 probability. When the probability of the packet drops increases then the throughput gets decreases. Indeed, we need to complement existing congestion control algorithms with some intelligent mechanisms that can differentiate whether a certain packet drop is because of congestion or channel error thus avoid unnecessary window reduction. In order to distinguish between packets drops, we have developed a k-NN based module to guess whether the packet drops are due to the congestion or any other reasons. After integrating this module with CUBIC algorithm, we have observed significant performance improvement.