Proactive Edge Caching with LSTM-based Popularity Prediction

In 5G and future mobile networks, employing data caching in the edge network devices shows promise for enhancing network efficiency, saving energy, and improving user experience. An essential challenge lies in deciding which content to cache in edge caches with limited capacities, aiming to maximize the cache-hit ratio. This task is complicated by the dynamic evolution of content popularities in both local and global domains. This study involves assessing content popularity scores using an LSTM (Long Short Term Memory) model, utilizing past download data across local and global scopes. These scores aid in formulating an optimal caching problem akin to a knapsack problem, with a solution derived from a genetic algorithm (GA). Through a numerical investigation, we demonstrate the effectiveness of the proposed prediction and caching algorithms under various popularity models. The findings indicate that the suggested LSTM-based proactive caching algorithm performs notably better compared to the benchmark algorithms, particularly with smaller cache capacities and more diverse contents.