AI-driven simulated DDoS attacks in OMNeT++ using SHAP-based explainability analysis

Research on Distributed Denial-of-Service (DDoS) attacks increasingly uses network simulators to generate controlled, secure, and reproducible experimental traffic. This study presents a diagnostic workflow for evaluating cross-environment transfer of machine-learning-based DDoS detection. eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Multilayer Perceptron (MLP), and majority-voting ensemble models were trained on the Canadian Institute for Cybersecurity Distributed Denial of Service 2019 (CICDDoS2019) Benign-versus-DDoS task and applied without retraining to labeled traffic generated using the Objective Modular Network Testbed in C++ (OMNeT++) / Internet Networking (INET) Framework. A repeated-split benchmark with nested feature subsets was used to assess robustness while minimizing leakage risk. The canonical eight-feature configurations achieved mean classification accuracies above 98% across all model families. Cross-environment performance, however, was strongly scenario-dependent. Normal and User Datagram Protocol (UDP) flood traffic were classified consistently by tree-based and hybrid models, whereas Domain Name System (DNS) amplification performance varied across model configurations. Critically, the principal eight-feature consensus models provided almost no overall DDoS detection, with an aggregate recall of only 3.6% across 8,300 simulated attack flows. For TCP connection exhaustion specifically, all 8,000 attack flows were missed. This demonstrates a substantial cross-environment transfer failure despite strong benchmark performance. SHapley Additive exPlanations (SHAP) analysis of the frozen eight-feature XGBoost model, complemented by an auxiliary 79-predictor benchmark-space analysis, identified features influencing predictions and generated testable hypotheses for refining simulated traffic without implying causal simulator mechanisms. Code, simulation configurations, provenance manifests, and derived evidence are publicly available to support reproducibility.

Authors

Institutions

Publication Details

Journal
Discover Computing
Published
2026-11-19
DOI
https://doi.org/10.1007/s10791-026-10731-8
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

AI-driven simulated DDoS attacks in OMNeT++ using SHAP-based explainability analysis

Abbass Nasser, Ahmad Hani El Fawal, Abir Amer, Guillaume Andrieux et al.
Discover Computing
Network Security and Intrusion Detection
article

AI-driven simulated DDoS attacks in OMNeT++ using SHAP-based explainability analysis

Abbass Nasser, Ahmad Hani El Fawal, Abir Amer, Guillaume Andrieux, Hadi Shour, Ali Mansour
article en

Abstract

Research on Distributed Denial-of-Service (DDoS) attacks increasingly uses network simulators to generate controlled, secure, and reproducible experimental traffic. This study presents a diagnostic workflow for evaluating cross-environment transfer of machine-learning-based DDoS detection. eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Multilayer Perceptron (MLP), and majority-voting ensemble models were trained on the Canadian Institute for Cybersecurity Distributed Denial of Service 2019 (CICDDoS2019) Benign-versus-DDoS task and applied without retraining to labeled traffic generated using the Objective Modular Network Testbed in C++ (OMNeT++) / Internet Networking (INET) Framework. A repeated-split benchmark with nested feature subsets was used to assess robustness while minimizing leakage risk. The canonical eight-feature configurations achieved mean classification accuracies above 98% across all model families. Cross-environment performance, however, was strongly scenario-dependent. Normal and User Datagram Protocol (UDP) flood traffic were classified consistently by tree-based and hybrid models, whereas Domain Name System (DNS) amplification performance varied across model configurations. Critically, the principal eight-feature consensus models provided almost no overall DDoS detection, with an aggregate recall of only 3.6% across 8,300 simulated attack flows. For TCP connection exhaustion specifically, all 8,000 attack flows were missed. This demonstrates a substantial cross-environment transfer failure despite strong benchmark performance. SHapley Additive exPlanations (SHAP) analysis of the frozen eight-feature XGBoost model, complemented by an auxiliary 79-predictor benchmark-space analysis, identified features influencing predictions and generated testable hypotheses for refining simulated traffic without implying causal simulator mechanisms. Code, simulation configurations, provenance manifests, and derived evidence are publicly available to support reproducibility.

Discover ComputingVol. 29(1)
Centre National de la Recherche Scientifique (FR), Holy Spirit University of Kaslik (LB), Modern University for Business and Science (LB), Institut d'Électronique et des Technologies du numéRique (FR), Laboratoire des Sciences et Techniques de l’Information de la Communication et de la Connaissance (FR), Nantes Université (FR)
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