Popularity Prediction Algorithm for Cache-enabled Networks

  • Huda S. Goian

Student thesis: Master's Thesis


Huda S. Goian, 'Popularity Prediction Algorithm for Cache-enabled Networks', M.Sc. Thesis, M.Sc. in Electrical and Computer Engineering, Department of Electrical and Computer Engineering, Khalifa University of Science and Technology, United Arab Emirates, June 2018. The proliferation of the mobile Internet and smart handheld devices, which leads to a variety of services at different levels of performance, is a major challenge for the fifth generation of wireless networks and beyond. Innovative solutions are needed to leverage recent advances in machine storage/memory, context awareness, and edge computing. Cache-enabled networks and techniques such as edge and proactive caching could help to reduce content delivery times and traffic congestion in wireless networks. Only a few contents are popular, accounting for the majority of viewers, so caching them reduces the latency and download time. However, given the dynamic nature of user behavior, the integration of popularity prediction into caching could help to prepare the content in advance, resulting in better network utilization and user satisfaction. This thesis project aims to design and develop a prediction model capable of accurately anticipating the next day videos popularity and selecting videos to be cached based on users' behaviors and preference. First, we propose a deep learning model which forecasts the following day demand. The model operates during off-peak hours, and it must be quick to complete predictions and decide contents locations before the next peak hour. Second, another algorithm is designed to handle new videos uploaded during the day. The algorithm is developed to cope with the inherited trade-off between accurate and early prediction. In more detail, this thesis starts with demonstrating the value of popularity prediction for caching networks and the next generation wireless networks (5G). It also provides a general framework for popularity based caching, and reviews the theoretical background of machine learning algorithms used for cache-enabled networks. In addition, it states the main shortcomings of each machine learning category before determining our algorithm. Then, the design and development of two prediction models: a day ahead video popularity predictor and new videos popularity predictor, are discussed and evaluated in term of accuracy and time requirements. Numerical results are provided in order to show the effectiveness of the two models. Indexing Terms: 5G, Cache-enabled networking, Deep learning, Popularity prediction, Proactive caching, Videos popularity
Date of AwardJun 2018
Original languageAmerican English
SupervisorYousof Al Hammadi (Supervisor)


  • 5G
  • Cache-enabled networking
  • Deep learning
  • Popularity prediction
  • Proactive caching
  • Videos popularity.

Cite this