Video is becoming one of the biggest traffic generating elements and the current progressive download approach has been found inefficient in terms of resource utilization. A machine learning based approach is proposed for the video streaming with DASH (Dynamic adaptive streaming over HTTP) as an underlying architecture to help client adapt to the changing streaming environment. The reason for using machine learning for adaptation is to make client learn about the environment that too in unsupervised manner. This way the redundant work of adaptation for similar network signatures is eliminated, therefore running time complexity is reduced. The results of our work showed improved QoE and efficient bandwidth utilization, (upto 68.5%), compared with the already present DASH algorithms.