Recently, it has been discovered that EEG signals have enormous potential to be used as biometric authentication. Although, its practical implementation is limited due to the intricate and dynamic nature of EEG signals. To overcome these challenges, we need to simplify the analysis and preserve the spatial attributes of the EEG signals. In this work, a methodology using an ensemble of Riemannian geometry and a genetic algorithm for EEG-based biometric authentication is devised. Here, the symmetric positive definite covariance matrices of the EEG signals are calculated and classified using the Minimum distance to the Riemannian Mean (MDRM) and the Tangent space LDA (TSLDA) classifier. Furthermore, NSGA-II is used to optimize the number of channels and to reduce the computational complexity. This study achieved an accuracy of 99.9% on average with all the datasets used. Multiple publicly available datasets are used to compare the proposed approach with existing methods. Results obtained show the efficacy of the proposed method. Friedman’s statistical test also supports the statistical significance difference between the proposed and existing methods.