Pattern Recognition for Adaptive Access Control Policy: Comparing GAIL and Decision Trees in Dynamic Organizational Structures

Access control manages who can access organizational resources and under what conditions. This study compares two machine learning techniques, Generative Adversarial Imitation Learning (GAIL) and Decision Trees (DT), for pattern recognition in Adaptive Access Control Policy (AACP) learning within dynamic organizational environments. The traditional access control systems are statically crafted, requiring manual policy updates if a change occurs, creating administrative burdens and security risks if the organization evolves. We evaluated both models to measure their generalization capabilities using a controlled experimental approach with an augmented Amazon Employee Access dataset across four organization scenarios. GAIL performed with 97.46% test accuracy compared to DT, achieved 91.16%. Furthermore, GAIL exhibited excellent generalization with a -0.07% validation-test gap compared to DT's 3.28%. These findings offer empirical guidance for organizations seeking to implement adaptive access control systems in evolving environments.