Related Experiment Video
Updated: Oct 30, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.1K
Improving the Performance of Machine Learning-Based Network Intrusion Detection Systems on the UNSW-NB15 Dataset
Soulaiman Moualla1, Khaldoun Khorzom1, Assef Jafar1
1Department of Telecommunication, Higher Institute for Applied Sciences and Technology, Damascus, Syria.
Computational Intelligence and Neuroscience
|July 5, 2021
Summary
This study introduces a novel machine learning-based network intrusion detection system (IDS) that enhances cybersecurity by improving detection rates and reducing false alarms. The system effectively addresses imbalanced data and identifies various cyber threats using advanced techniques.
Area of Science:
- Computer Science
- Cybersecurity
- Machine Learning
Background:
- Traditional firewalls and intrusion detection systems (IDS) require constant database updates to counter evolving cyber threats.
- Machine learning offers a path toward more robust network defense systems with higher detection accuracy and fewer false alarms.
Purpose of the Study:
- To develop a novel, dynamically scalable, multiclass machine learning-based network intrusion detection system (IDS).
- To enhance network security by improving the detection rates and reducing the false alarm rates of cyberattack detection.
Main Methods:
- The proposed IDS utilizes supervised machine learning, beginning with the Synthetic Minority Oversampling Technique (SMOTE) for class imbalance.
- Feature selection is performed using the Gini Impurity criterion with an Extremely Randomized Trees Classifier.
- A pretrained Extreme Learning Machine (ELM) model acts as a binary classifier for each attack type (One-Versus-All), with outputs feeding into a fully connected layer and a logistic regression layer for final classification.
Main Results:
- The novel IDS demonstrates superior performance compared to existing methods on the UNSW-NB15 dataset.
- Key performance metrics including accuracy, false alarm rate, Receiver Operating Characteristic (ROC) curves, and Precision-Recall Curves (PRCs) show significant improvements.
Conclusions:
- The proposed machine learning-based IDS offers a robust and effective solution for modern network security challenges.
- The system's dynamic scalability and advanced feature selection/classification methods contribute to its high performance in detecting cyberattacks.
