SPARCQ: Enhancing Scalability and Adaptability of Proactive Edge Caching Through Q-Learning

The exponential growth of network traffic and data-intensive applications demands innovative solutions to manage data efficiently and ensure high-quality user experiences. Proactive edge caching has become a crucial technique for enhancing network performance by predicting and pre-storing content closer to users before access. Accurate prediction models, such as Long Short-Term Memory (LSTM) networks, are crucial for effective proactive caching. However, these models rely on carefully tuned hyperparameters to maintain predictive accuracy, and manual tuning is impractical in dynamic and diverse network environments, limiting scalability and adaptability. To overcome these challenges, we propose a novel framework, SPARCQ, that leverages Q-learning, a reinforcement learning algorithm, to automate hyperparameter tuning for LSTM-based prediction models. By dynamically adjusting hyperparameters, our approach ensures accurate predictions, improving caching efficiency and adaptability. Using the MovieLens dataset, we achieve an average improvement of 8% in cache hit ratios compared to baseline models, including popularity-based and untuned models. Additionally, our framework demonstrates scalability and robustness across geographically distributed regions, consistently adapting to diverse and evolving data patterns.