The present disclosure provides a system and methods for determining and reserving an optimized amount of required cloud instance resources in order to provide quality streaming to viewers while also reducing overhead costs. For content providers, the choice of appropriate cloud location to transcode the live streams and appropriate number of video representations to serve the current viewership is a key in minimizing cost. However, this task cannot typically be done on the fly due to the properties of live streaming characterized by its dynamic, heterogeneity and wideness. The provided system helps predict the incoming load and the exact number of resources to allocate at each cloud site to help prevent the content provider from over-provisioning resources, leading to an extra cost, or under-provisioning resources, leading to viewers' dissatisfaction.