Integration of simulated annealing into pigeon inspired optimizer algorithm for feature selection in network...

Area of Science:

• Cybersecurity
• Network Engineering
• Artificial Intelligence

Background:

• 5G networks face increased traffic and complex patterns due to numerous devices.
• Traditional intrusion detection systems struggle with high-dimensional and complex network data.
• This leads to complex models, reduced accuracy, and prolonged detection times.

Purpose of the Study:

• To propose an improved feature selection algorithm (SABPIO) for network intrusion detection.
• To address the challenges of high dimensionality and complexity in 5G network traffic.
• To enhance the accuracy and efficiency of intrusion detection systems.

Main Methods:

• Pre-processing raw network data using one-hot encoding and standardization.
• Employing the Simulated Annealing and Population Decay Factor within SABPIO for feature selection.
• Evaluating the selected features using decision trees and random forest classifiers on UNSW-NB15, NLS-KDD, and CIC-IDS-2017 datasets.

Main Results:

• SABPIO effectively identifies the most indicative subset of network traffic features.
• The method significantly reduces system training duration (3.2x) and testing times (0.3x).
• Achieved F1-score improvements of 1.21%-2.19% over CPIO and 1.79%-4.52% over XGBoost.

Conclusions:

• SABPIO offers a rational and effective approach to feature selection for network intrusion detection.
• The algorithm enhances detection rates and reduces computational overhead in 5G environments.
• SABPIO demonstrates superior performance compared to existing methods in improving intrusion detection metrics.