Related Experiment Video
Updated: Jul 16, 2026
05:30
Large Scale Energy Efficient Sensor Network Routing Using a Quantum Processor Unit
Published on: September 8, 2023
MM-NIDS: A Novel Multimodal Ensemble Fusion Network Intrusion Detection System Using Numeric, Text, Graph, and
Samar AboulEla1, Rasha Kashef1
1Electrical, Computer, and Biomedical Engineering, Toronto Metropolitan University, Toronto, ON M5B 2K3, Canada.
Sensors (Basel, Switzerland)
|July 15, 2026
Summary
This study introduces MM-NIDS, a multimodal fusion framework for network intrusion detection using NetFlow data. It enhances cybersecurity for IoT and critical infrastructure by improving the detection of evolving cyber threats.
Area of Science:
- Cybersecurity and Network Engineering
- Artificial Intelligence and Machine Learning
Background:
- The proliferation of Internet of Things (IoT) devices increases exposure to sophisticated cyber threats.
- Traditional intrusion detection systems struggle with resource limitations, evolving attacks, and diverse traffic patterns in modern networks.
Purpose of the Study:
- To introduce MM-NIDS, a novel multimodal fusion framework for NetFlow-based intrusion detection.
- To enhance the robustness and adaptability of intrusion detection systems for complex digital environments.
Main Methods:
- Developed a multimodal fusion framework (MM-NIDS) combining numerical, textual, graph-based, and quantum-inspired NetFlow data representations.
- Utilized transformer-based architectures (FT-Transformer, ELECTRA-Small) for modeling each data representation.
- Integrated predictions using five post hoc fusion strategies, including MLP and XGBoost meta-fusion.
Main Results:
- The text-based model (M2) demonstrated the highest individual performance.
- Fusion strategies offered modest, dataset-dependent improvements, particularly for underrepresented attacks.
- NetFlow features proved effective for volumetric and scan-based attacks but less so for stealthy threats.
Conclusions:
- MM-NIDS shows potential for deployment in critical infrastructure, industrial IoT, and smart environments.
- Further research should incorporate semantic or payload-level features to improve detection of evasive threats.