Research on Adaptive 1DCNN Network Intrusion Detection Technology Based on BSGM Mixed Sampling.

Area of Science:

• Cybersecurity
• Machine Learning
• Data Science

Background:

• Network security is threatened by increasing cyberattacks.
• Imbalanced datasets, with scarce attack data, hinder classifier performance.
• Optimizing convolutional neural network (CNN) parameters, like convolutional kernels, is challenging.

Purpose of the Study:

• To address class imbalance in network intrusion detection.
• To automatically optimize CNN parameters for enhanced detection.
• To improve the detection rate of minority attack classes.

Main Methods:

• A hybrid sampling technique, Borderline-SMOTE and Gaussian Mixture Model (BSGM), was developed.
• Quantum Particle Swarm Optimization (QPSO) was used to determine optimal convolutional kernels for 1DCNN.
• The BSGM-QPSO-1DCNN method was evaluated on the KDD99 dataset against benchmark models.

Main Results:

• BSGM-QPSO-1DCNN achieved high accuracy rates of 99.93% (binary) and 99.94% (multi-class).
• Precision for minority classes R2L and U2R improved by 68% and 66%, respectively.
• The proposed method outperformed five other intrusion detection models.

Conclusions:

• BSGM-QPSO-1DCNN is an effective solution for imbalanced data in network intrusion detection.
• The hybrid approach significantly enhances the detection of rare but critical network attacks.
• This research offers a robust method for improving cybersecurity defenses.