In Federated edge learning (FEEL), energy-constrained devices at the network\nedge consume significant energy when training and uploading their local machine\nlearning models, leading to a decrease in their lifetime. This work proposes\nnovel solutions for energy-efficient FEEL by jointly considering local training\ndata, available computation, and communications resources, and deadline\nconstraints of FEEL rounds to reduce energy consumption. This paper considers a\nsystem model where the edge server is equipped with multiple antennas employing\nbeamforming techniques to communicate with the local users through orthogonal\nchannels. Specifically, we consider a problem that aims to find the optimal\nuser's resources, including the fine-grained selection of relevant training\nsamples, bandwidth, transmission power, beamforming weights, and processing\nspeed with the goal of minimizing the total energy consumption given a deadline\nconstraint on the communication rounds of FEEL. Then, we devise tractable\nsolutions by first proposing a novel fine-grained training algorithm that\nexcludes less relevant training samples and effectively chooses only the\nsamples that improve the model's performance. After that, we derive closed-form\nsolutions, followed by a Golden-Section-based iterative algorithm to find the\noptimal computation and communication resources that minimize energy\nconsumption. Experiments using MNIST and CIFAR-10 datasets demonstrate that our\nproposed algorithms considerably outperform the state-of-the-art solutions as\nenergy consumption decreases by 79% for MNIST and 73% for CIFAR-10 datasets.\n