Big Data Driven Predictive Caching at the Wireless Edge

Abstract

The effective delivery of high bandwidth and high quality-of-experience mobile services to increasingly mobile users has become a major challenge for mobile carriers. While content-delivery networks improve the effectiveness of fixed networks, caching at the edge of mobile networks is challenging due to heterogeneity in service and content usage patterns, and variations in user mobility patterns. To address this challenge, machine learning-based predictive analytics could be used to anticipate user behaviors and content request patterns, and pre-fetch content with high expected hit rates to wireless edge caches close to the users. Although recent research has shifted to focus on machine learning-based caching techniques, the optimal strategy and performance of predictive wireless edge caching have yet to be fully investigated. Here, we first develop a generic predictive algorithm to predict the daily trajectories and the service usage patterns of mobile users. We then investigate the critical factors that affects the users' prediction accuracy based on real network datasets. Next, we propose a new caching technique that utilizes our predictive algorithm to anticipate user requests, and we design a simulation to comprehensively generate the trajectories and service content requests of users in a mobile network. We then establish base cases to first show that the predictive caching strategy outperforms the non-predictive method.