Optimizing feature selection with random reversal and adaptive Gaussian based Dung beetle optimizer for intrusion detection system in IoT

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989100%3A27240%2F25%3A10260263" target="_blank" >RIV/61989100:27240/25:10260263 - isvavai.cz</a>

Result on the web

<a href="https://www.nature.com/articles/s41598-025-29278-7" target="_blank" >https://www.nature.com/articles/s41598-025-29278-7</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1038/s41598-025-29278-7" target="_blank" >10.1038/s41598-025-29278-7</a>

Result language

angličtina

Original language name

Optimizing feature selection with random reversal and adaptive Gaussian based Dung beetle optimizer for intrusion detection system in IoT

Original language description

The Internet of Things (IoT) is an emerging, promising technology developed with the objective of establishing global connectivity among devices. IoT is highly susceptible to malicious attacks, owing to its resource-constrained architecture, insecure wireless communication, diverse device ecosystems, and the vast volume of sensor data transmitted over networks. An effective Intrusion Detection System (IDS) is essential to address these security concerns. However, challenges such as irrelevant features and poor class separability complicate its development. This research proposes a novel IDS by introducing an Improved Random Reversal Learning (IRRL) and Dimensional Adaptive Gaussian Variation (DAGV)-based Dung Beetle Optimizer (RGDBO) for optimal feature selection, enhancing exploration, and avoiding premature convergence. For classification, a Convolutional Neural Network (CNN) integrated with CosFace and ArcFace loss functions, termed CACNN, is employed to enhance intrusion classification through more efficient discrimination among classes. The combined RGDBO-CACNN framework is evaluated on three benchmark datasets: UNSW-NB15, NSL-KDD, and CICIDS-2017, using accuracy, recall, precision, and F1-score as performance metrics. A comparative analysis of existing methods, including GA-FR-CNN, GTO-BSA, and BMRF-RF, demonstrates the superiority of the proposed model, with RGDBO-CACNN achieving an accuracy of 99.999% on the UNSW-NB15 dataset.

Czech name

—

Czech description

—

Type

J_imp - Article in a specialist periodical, which is included in the Web of Science database

CEP classification

—

OECD FORD branch

20200 - Electrical engineering, Electronic engineering, Information engineering

Project

—

Continuities

O - Projekt operacniho programu

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Name of the periodical

Scientific Reports

ISSN

2045-2322

e-ISSN

2045-2322

Volume of the periodical

15

Issue of the periodical within the volume

1

Country of publishing house

GB - UNITED KINGDOM

Number of pages

25

Pages from-to

nestránkováno

UT code for WoS article

001651215100002

EID of the result in the Scopus database

—