CacheHitPredictor: A Federated Learning Approach to Edge Cache Hit Prediction in URLLC Systems

Efficient edge caching is vital for meeting the stringent latency and reliability demands of Ultra-Reliable Low Latency Communication (URLLC) in 5G and beyond. This paper proposes <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CacheHitPredictor</i>, a federated learning-based framework for predictive edge caching that preserves data privacy while handling client heterogeneity and dynamic user behavior. A URLLC scenario was simulated using CloudSimPlus to generate realistic user request traces. The resulting dataset was used to train a federated neural network model via the Flower framework, with varying Dirichlet factors to evaluate data heterogeneity. The trained model was deployed via a Flask REST API and integrated with the CloudSimPlus simulation to enable online cache decision-making. Experimental results demonstrate that the proposed model performs on par with LFU under skewed workloads, underscoring the practical potential of federated learning to enable adaptive and efficient caching in URLLC edge environments.