Work

ImprovedRFECV with L1 and L2 regularization and decision tree for automated feature selection in SDN-IoT threat detection and threat classification

Year: 2026

Type: article

Abstract: The rising adoption of Internet of Things (IoT) technologies along with the integration of Software Defined Networking (SDN) has led to significant advances in programmability, flexibility, as well as the management of networks. At the same time, it has raised a host of security concerns, thereby creating a pressing need for the development of sophisticated threat detection as well as threat classification systems. Machine Learning (ML) has gained significant traction in the development of threat detection as well as threat classification systems due to the rising demand for automated anomaly detection systems. In recent years, there has been a surge in the development of ML-based threat detection as well as threat classification systems, yet there has been a significant deficiency in the development of sophisticated systems in the context of important stages such as feature selection as well as classifier development in binary classification (BCT) as well as multi-class classification (MCT). This study aims to fill in the gaps by evaluating the effect of NIA in ML systems in the context of binary classification as well as multi-class classification, as well as an improved method of feature selection known as RFECV along with a Decision Tree classifier as well as an integration of L1 as well as L2 regularizes. For BCT, the full TD-TC pipeline achieves 91.07% (DT) and 90.82% (RF) balanced accuracy on IoTID20, and 99.96% (DT and RF) on InSDN. For MCT, balanced accuracy reaches 82.53% (DT) and 80.02% (RF) on IoTID20, and 93.52% (DT) and 94.74% (RF) on InSDN. Per-batch testing time ranges from 0.012 to 1.18 s across all eight configurations, supporting the framework’s potential for real-time SDN-IoT intrusion detection.

Source: Scientific Reports

DOI: 10.1038/s41598-026-67852-9

Authors: Teoh Ren Nee, Tan Saw Chin, Farah Kamila Binti Yahya, Jimmy Herawan Moedjahedy

Institutions: Multimedia University, Klabat University

Language: English

Topic: Network Security and Intrusion Detection

Subfield: Computer Networks and Communications

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $2,690

Funders: Multimedia University, Ministry of Higher Education, Malaysia