Signal classification via electroencephalography (EEG) is essential for smart healthcare systems and human-machine interaction systems. Centralized machine learning approaches are problematic because they require collecting sensitive neural data from multiple participants, which raises seriously privacy concerns. This paper proposes a privacy-preserving federated learning approach for classifying EEG signals in a distributed manner. The framework uses Convolutional Neural Networks (CNN) and two federated aggregation techniques: Federated Averaging (FedAvg) and Federated Proximal (FedProx) to allow the collaborative training of models without revealing individual raw EEG datasets. The approach is tested on the PhysioNet EEG Motor Imagery dataset with 109 subjects. The performance of the federated framework is comparable to the centralized framework, thus, successfully preserving privacy. More specifically, the model based on FedProx, in a specified Interval of Data (IID), classification achieved 90.36%, exceeding many current federated EEG learning methods. The findings reinforce the value of federated optimization methods for confidential distributed learning over diverse EEG datasets.