Network Anomaly Detection via a Dynamic-Aware Online Transfer Learning.

Saved in:

Bibliographic Details
Title:	Network Anomaly Detection via a Dynamic-Aware Online Transfer Learning.
Authors:	Gu, Huanan¹ 1113965312@qq.com, Feng, Quanxi² fqx9904@163.com
Source:	IAENG International Journal of Computer Science. Apr2026, Vol. 53 Issue 4, p1509-1522. 14p.
Subjects:	Anomaly detection (Computer security), Online data processing, Change-point problems, Internet security, Feature selection, Knowledge transfer
Abstract:	In dynamic network environments, concept drift poses a critical challenge to streaming data analysis. Existing online transfer learning methods, constrained by static weight allocation, fixed feature spaces, and delayed drift detection, often struggle to achieve a balance between stability and adaptability. To overcome these limitations, this paper proposes a Dynamic-Aware Online Transfer Learning (DAOTL) model that integrates three synergistic mechanisms to enhance adaptability under evolving data distributions. First, a Dynamic Adaptive Weight Decay (DAWD) mechanism combines instantaneous classifier performance with lifecycle-aware decay, enabling the model to retain historical knowledge while rapidly adapting to abrupt drifts. Second, an Incremental Feature Optimization Framework (IFOF) dynamically adjusts feature subsets within a sliding window to alleviate performance degradation caused by sudden shifts in feature space. Third, a drift-event-triggered mechanism monitors ensemble prediction confidence and replaces fixed-window updates with data-driven adaptations, enabling timely responses to drift events. Experimental evaluations on NSL-KDD and CSE-CIC-IDS2018 datasets, which are widely used benchmarks for network anomaly detection, demonstrate that DAOTL improves multi-class classification accuracy by 6.08% and 3.77%, respectively, compared to baseline models, validating its effectiveness in achieving both adaptability and efficiency in dynamic environments. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	In dynamic network environments, concept drift poses a critical challenge to streaming data analysis. Existing online transfer learning methods, constrained by static weight allocation, fixed feature spaces, and delayed drift detection, often struggle to achieve a balance between stability and adaptability. To overcome these limitations, this paper proposes a Dynamic-Aware Online Transfer Learning (DAOTL) model that integrates three synergistic mechanisms to enhance adaptability under evolving data distributions. First, a Dynamic Adaptive Weight Decay (DAWD) mechanism combines instantaneous classifier performance with lifecycle-aware decay, enabling the model to retain historical knowledge while rapidly adapting to abrupt drifts. Second, an Incremental Feature Optimization Framework (IFOF) dynamically adjusts feature subsets within a sliding window to alleviate performance degradation caused by sudden shifts in feature space. Third, a drift-event-triggered mechanism monitors ensemble prediction confidence and replaces fixed-window updates with data-driven adaptations, enabling timely responses to drift events. Experimental evaluations on NSL-KDD and CSE-CIC-IDS2018 datasets, which are widely used benchmarks for network anomaly detection, demonstrate that DAOTL improves multi-class classification accuracy by 6.08% and 3.77%, respectively, compared to baseline models, validating its effectiveness in achieving both adaptability and efficiency in dynamic environments. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X