Boosted-3R: Towards a Novel Framework for Inferring ABAC Policies

In the realm of Identity and Access Management, access control mechanisms are established to prevent unauthorized entry, ensuring that only authorized entities receive permission to access resources. Among access control models, AttributeBased Access Control (ABAC) stands out by considering diverse attributes of users, resources, and the environment, enabling finegrained control and dynamic policy implementation. However, implementing and migrating to ABAC model poses challenges, particularly in large organizational contexts. This paper introduces Boosted-3R, an innovative approach that employs federated machine learning and CatBoost, an advanced algorithm tailored for categorical feature handling, to derive ABAC rules from access logs using association Rule mining, Reliability verification and Redundancy deletion. Utilizing pattern mining and categorical feature classification, our framework extracts reliable rules from access logs, enabling the prediction of access decisions for new requests. The proposed approach is evaluated using a real dataset from Amazon, and experimental results showcase an accuracy score of 96.7%.