Currently, Video on Demand (VoD) is the heaviest data-driven service among various multimedia services. Therefore, predicting the popularity of mul-timedia contents, particularly video contents, in order to supply their proactive caching is a crucial issue. Mobile Edge Computing is proposed as a general solution for the upcoming 5G-based networks to allow computation and storage capabilities at the edge of Radio Access Network. Henceforth, applications and services can be deployed near the end-user, improving the backhaul resources utilization, energy consumption, and users Quality of Experience thanks to the reduction of traffic latency. Therefore, the pivotal problem addressed in this paper is to find the most appropriate prediction methods from a large set of predefined methods, able to forecast, with high accuracy, the multimedia content popularity. A real YouTube dataset was used for simulations. The proposed Recommendation System framework is very flexible because it can be extended to different time-series prediction contexts. It also displays good scalability and can be dynamically adapted for analyzing huge sets of time-series with large sets of predictors.