Optimized intrusion detection using particle swarm optimization and neural networks in simulated and physical network environments

This method article presents a reproducible framework for constructing and evaluating an intrusion detection system using optimized machine learning workflows. Network traffic was collected by implementing two independently controlled setups: a software-based network model and a small-scale real hardware network, with packet traces recorded using Wireshark. These traces were merged with widely used reference datasets to ensure coverage of diverse attack behaviors. Feature dimensionality was reduced using Particle Swarm Optimization, enabling efficient learning while limiting redundant network attributes. The refined feature sets were subsequently used to train multiple classification models, including tree-based, distance-based, and neural learning approaches. Experimental evaluation shows that the neural model trained on PSO-selected features delivers the most stable detection behavior, combining strong classification performance with a consistently low false alert rate, thereby validating the proposed method for intrusion analysis studies. • Generation of packet-level datasets from independently built virtual and physical network environments • Optimization-driven feature selection using Particle Swarm Optimization to reduce redundancy • Cross-model evaluation of optimized classifiers, demonstrating superior reliability of the PSO-enhanced neural approach.