In this paper we study the end-to-end complete feedback collection (ECFC) problem in large scale multicast applications. We consider the case where each receiver is expected to send feedback in a timely manner without causing implosion at the source site. To address the scalability problem and improve timely feedback collection, we introduce the use of clustering algorithms for feedback collection. Our simulation based comparisons show that the clustering based approaches outperform the existing pure (without clustering) multi-round probabilistic and pure (without clustering) delayed feedback collection approaches both in terms of collection delay and message overhead.