Intrusion detection system based on machine learning using least square support vector machine

Area of Science:

• Cybersecurity
• Machine Learning
• Network Intrusion Detection

Background:

• Traditional intrusion detection systems (IDS) struggle with the scale and complexity of big data, leading to inaccuracies and false predictions.
• Machine learning (ML) offers improved performance for IDS, but conventional ML-based systems face challenges in scalability and high dimensionality, often declining in accuracy with irrelevant features.
• Minimizing false positives and improving the efficiency of ML-based IDS are critical for effective real-time network protection.

Purpose of the Study:

• To develop a machine learning-based network intrusion detection system that enhances classification accuracy and reduces false predictions.
• To address the scalability and efficiency challenges of traditional ML-based IDS in big data environments.
• To introduce a supervised framework utilizing a quantum-inspired least square support vector machine (LS-SVM) with exhaustive feature selection.

Main Methods:

• Employed an exhaustive feature selection algorithm to identify the optimal subset of features for maximum classification accuracy.
• Utilized a quantum-inspired least square support vector machine (LS-SVM) classifier within a supervised learning framework.
• Tuned model hyperparameters based on the selected features to maximize predictive performance and minimize training time.

Main Results:

• Achieved high detection accuracy: 99.3% on NSL-KDD, 99.5% on CIC-IDS-2017, and 93.3% on UNSW-NB15.
• Demonstrated excellent precision (1.00 for CIC-IDS-2017 and UNSW-NB15) and recall (up to 1.00 for CIC-IDS-2017).
• Exhibited competitive testing times (e.g., 1.0s for CIC-IDS-2017) and minimal training times compared to other models, indicating suitability for real-time applications.

Conclusions:

• The proposed LS-SVM-based intrusion detection system effectively enhances classification accuracy and reduces false predictions, outperforming conventional methods.
• The integration of exhaustive feature selection and quantum-inspired LS-SVM provides a scalable and efficient solution for big data network security.
• The model's robust performance and efficiency make it a promising candidate for real-time intrusion detection applications.