Telemetry Dataset for Adaptive Federated Reinforcement Learning in Edge–Cloud Systems

This dataset contains telemetry traces collected from a heterogeneous edge–cloud federated learning environment deployed on AWS EC2. The dataset is designed to support research in adaptive federated reinforcement learning (AFRL), particularly for telemetry-driven client participation and computation offloading. The dataset consists of approximately 130,000 samples collected from 10 heterogeneous edge clients under controlled system conditions, including normal operation, high CPU load, disk load, and network fluctuation scenarios. Each telemetry sample includes 15 system-level features capturing compute, memory, storage, and network behaviour. These include CPU utilisation, memory usage, system load, disk throughput, network throughput, and network quality metrics such as latency, jitter, and packet loss. The dataset is used to construct compact state representations for reinforcement learning policies and to evaluate adaptive offloading, client selection, and system coordination strategies in federated learning. This dataset supports the experimental evaluation presented in the associated research work on telemetry-driven dual-policy federated reinforcement learning (TR-DP-AFRL). The full implementation, configuration files, and experiment scripts are available at:https://github.com/DrSaanjannaYuvaraj/tr-dp-afrl-edge-cloud-artifact