Institutional Repository of Vilnius University
Prieigos kontrolės plėtojimas DevOps įrankiuose: laiku pagrįstų taisyklių įgyvendinimas /
Abstract
dc:descriptionThis paper investigates the application of machine learning to the detection of security anomalies in access control audit logs of Kubernetes clusters. The goal of this work is to develop a technological solution that combines timelimited JustInTime authorization with dynamic behavior assessment, enabling the detection of anomalies in real time and the automatic revocation of access. The thesis developed a policy engine system that implements JIT access control in Kubernetes clusters with OIDC authentication via the Keycloak identity provider and a Redis session store. A centralized auditing mechanism was implemented to record structured data on user actions during sessions, including GeoIP enrichment. A synthetic set of audit records (87339 events, 5 anomaly types) was generated for the machine learning experiment. XGBoost and LightGBM gradient boosting models were trained and compared. Both models achieved high anomaly detection results. XGBoost features lower execution latency and a smaller model size, making it recommended for realtime integration. The trained models were exported to ONNX format and integrated into the policy engine for automatic access revocation upon anomaly detection.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Dvornikovas, Markas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308336708&prefLang=en_US