Adaptive user behavior baseline construction under concept drift for insider threat detection systems

Insider threat detection systems based on user behavior analysis (UEBA) face the problem of concept drift — natural changes in employee activity due to job transitions, project changes, or work schedules. Ignoring drift leads to a sharp increase in false positives, while too rapid adaptation may miss a real attack. This paper proposes an adaptive baseline construction method that not only tracks behavior changes but also automatically distinguishes between legitimate drift and insider anomaly. The method combines Exponentially Weighted Moving Average (EWMA) for smooth profile parameter updating and the ADWIN algorithm for detecting significant change points. The key novelty lies in a two-stage classification procedure for detected changes: first, the persistence of the deviation over time is checked, then the consistency of the new behavior with profiles of employees in the same role is verified. Unlike known approaches, the proposed method uses role context to validate drift, reducing false positives without losing sensitivity to real threats. Experiments on the publicly available CMU-CERT r4.2 dataset show that the proposed method reduces the false positive rate by 37 % compared to a fixed profile and by 19 % compared to a fixed-width sliding window, while maintaining attack detection recall at 0.91. Sensitivity analysis, statistical significance evaluation, and comparison with modern drift detection methods are performed. The obtained results can be used in building industrial UEBA systems operating in dynamic corporate environ-ments.

Authors

Institutions

Publication Details

Journal
Modeling of systems and processes
Published
2026-11-13
DOI
https://doi.org/10.12737/2219-0767-2026-19-3-83-94
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Adaptive user behavior baseline construction under concept drift for insider threat detection systems

Sergey Saharov, Vasiliy Mal'cev, Aleksandr Achkasov, A. Yagodkin
Modeling of systems and processes
Network Security and Intrusion Detection
article

Adaptive user behavior baseline construction under concept drift for insider threat detection systems

Sergey Saharov, Vasiliy Mal'cev, Aleksandr Achkasov, A. Yagodkin
article en

Abstract

Insider threat detection systems based on user behavior analysis (UEBA) face the problem of concept drift — natural changes in employee activity due to job transitions, project changes, or work schedules. Ignoring drift leads to a sharp increase in false positives, while too rapid adaptation may miss a real attack. This paper proposes an adaptive baseline construction method that not only tracks behavior changes but also automatically distinguishes between legitimate drift and insider anomaly. The method combines Exponentially Weighted Moving Average (EWMA) for smooth profile parameter updating and the ADWIN algorithm for detecting significant change points. The key novelty lies in a two-stage classification procedure for detected changes: first, the persistence of the deviation over time is checked, then the consistency of the new behavior with profiles of employees in the same role is verified. Unlike known approaches, the proposed method uses role context to validate drift, reducing false positives without losing sensitivity to real threats. Experiments on the publicly available CMU-CERT r4.2 dataset show that the proposed method reduces the false positive rate by 37 % compared to a fixed profile and by 19 % compared to a fixed-width sliding window, while maintaining attack detection recall at 0.91. Sensitivity analysis, statistical significance evaluation, and comparison with modern drift detection methods are performed. The obtained results can be used in building industrial UEBA systems operating in dynamic corporate environ-ments.

Modeling of systems and processes
Voronezh State University of Forestry and Technologies (RU)
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon