SPARCQ: Enhancing Scalability and Adaptability of Proactive Edge Caching Through Q-Learning
Shruti Lall, Johan de Clercq, Nelishia Pillay, Bodhaswar T Maharaj · IEEE Access · 2025
The exponential growth of network traffic and data-intensive applications necessitates innovative solutions for efficient data management and high-quality user experiences. Proactive edge caching addresses this challenge by predicting and caching content closer to users. To enable accurate prediction, models such as Long Short-Term Memory (LSTM) networks are commonly used to learn temporal content request patterns. However, these models require carefully tuned hyperparameters to maintain accuracy, and manual tuning is both impractical and not scalable in dynamic network environments. To overcome this, we propose SPARCQ, a reinforcement learning-based framework that leverages Q-learning to automate hyperparameter optimization for LSTM-based prediction models. Our approach dynamically adapts to evolving content demand, improving both scalability and predictive accuracy. Using theMovieLensdataset, SPARCQ achieves an average 8% improvement in cache hit ratios, with gains of up to 22% in certain scenarios. Additionally, our framework remains robust across different cache parameters, maintaining high performance under varying caching conditions. While evaluated on LSTMs, SPARCQ is model-agnostic and can be extended to other predictive models, paving the way for broader applications in network optimization.