Machine-learning-based optical spectrum feature analysis for DoS attack detection in IP over optical networks - Research Article

Aug 3, 2026

This research presents a machine-learning approach to detect Denial of Service (DoS) attacks in software-defined IP over optical networks by analyzing optical spectrum features. The method automatically processes optical spectrum data to infer network security status and identify potential DoS events. Numerical simulations and experimental trials were conducted to build optical spectrum datasets. Performance of XGBoost, LightGBM, and a BP neural network was compared, achieving detection accuracy above 97%, with BP reaching 99.55% in simulations and 99.74% in experiments.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Background and motivation

Related work and limitations

Proposed scheme and evaluation

Optical-layer DoS detection via spectrum features

Simulation and experimental validation

Machine learning model comparison

Research Article Vol. 32, No. 3/29 Jan 2024/Optics Express 3793 Machine-learning-based optical spectrum feature analysis for DoS attack detection in IP over optical networks XIAOXUE GONG , 1,2 YANG LEI , 1,2 QIHAN ZHANG , 1,2 LU GAN , 3 XU ZHANG , 1,2,4 AND LEI GUO1,2 1 School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China 2 Institute of Intelligent Communications and Network Security, Chongqing University of Posts and Telecommunications, Chongqing 400065, China 3Dept. of Electronic and Electrical Engineering, Brunel University, London UB8 3PH, UK [4](4 zhangxu@cqupt.edu.cn)[ zhangxu@cqupt.edu.cn](4 zhangxu@cqupt.edu.cn) Abstract: In this paper, we introduce a novel approach for detecting Denial of Service (DoS) attacks in software-defined IP over optical networks, leveraging machine learning to analyze optical spectrum features. This method employs machine learning to automatically process optical spectrum data, which is indicative of network security status, thereby identifying potential DoS attacks. To validate its effectiveness, we conducted both numerical simulations and experimental trials to collect relevant optical spectrum datasets. We then assessed the performance of three machine learning algorithms XGBoost, LightGBM, and the BP neural network in detecting DoS attacks. Our findings show that all three algorithms demonstrate a detection accuracy exceeding 97%, with the BP neural network achieving the highest accuracy rates of 99.55% and 99.74% in simulations and experiments, respectively. This research not only offers a new avenue for DoS attack detection but also enhances early detection capabilities in the underlying optical network through optical spectrum data analysis. © 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement 1. Introduction With the booming development of various emerging network services, future-proof IP over optical networks is facing unprecedented opportunities and challenges [1–3] . Typically, IP over optical networks integrate IP and optical network technologies, where IP protocols are used to encapsulate data, while optical fiber is employed as a transmission medium to achieve high-speed transmission and exchange of data. Although IP over optical networks provides convenient service to users, they also suffer from serious network security threats. DoS attacks have become one of the main threats to network security due to their simple and crude attack mode, as well as diverse attack methods [4,5] . In general, the defense approaches for DoS attacks include three aspects: DoS attack detection, DoS attack source tracking, and DoS attack packet filtering. Among them, DoS attack detection is the foundation and premise of attack source tracking and attack packet filtering. Given the rising prevalence of transient DoS attacks, the need for fast and accurate detection methods has become an urgent priority in the field of network security research. Current DoS attack detection strategies primarily focus on packet-level analysis in the IP network layer [6–8], examining network traffic or highly processed packet datasets. While these methods, which analyze network traffic characteristics and packet contents, aid in detecting DoS attacks, they suffer from slow detection speeds. This sluggishness stems from their reliance on data obtained through photoelectric conversion, hindering their effectiveness against transient DoS attacks. Moreover, as IP and optical networks continue to merge and evolve rapidly, shifting \#513504 [https://doi.org/10.1364/OE.513504](https://doi.org/10.1364/OE.513504) Journal © 2024 Received 27 Nov 2023; revised 29 Dec 2023; accepted 3 Jan 2024; published 19 Jan 2024 Research Article Vol. 32, No. 3/29 Jan 2024/Optics Express 3794 more network functions to the optical layer via software-defined approaches is emerging asa tren

FAQs

How does the proposed method detect DoS attacks in IP over optical networks?

It uses machine learning to mine and process correlations in optical spectrum data, inferring whether the network security state indicates a DoS attack.

Which machine learning algorithms are evaluated in the study?

The study evaluates XGBoost, LightGBM, and a BP neural network, comparing their detection performance on optical spectrum-derived features.

What datasets and validation methods are used to test the approach?

The authors conduct numerical simulations using DoS traffic and normal traffic derived from a DARPA 1998 dataset, and they also run experimental trials by launching DoS attacks and monitoring spectrum data in real time.

Aug 3, 2026

Upload by Emma Wilson