Embodiments of the present invention provide a system for generative adversarial network training and feature extraction for biometric authentication. The system collects electronic biometric data of a user from one or more data sources, and stores the collected electronic biometric data as a biometric user account for the user in a personal NoSQL database library associated with the user. A generative adversarial neural network system then determines improved biometric feature selection and improved model refinements for existing biometric authentication models based on the biometric account for the user in the personal library associated with the user. The system can then determine user exposure levels for different authentication channels, including certain biometric authentication channels. A custom adversarial strategy for general adversarial network attacks is then established based on the user exposure levels to generate a biometric authentication process that is more accurate and secure.