Enhanced Intrusion Detection in Wireless Sensor Networks Using PSO-Optimised Machine Learning and Deep Learning Models on Protocol-Specific Datasets

Abstract

The effectiveness of Machine Learning (ML) and Deep Learning (DL) models for intrusion detection in Wireless Sensor Networks (WSNs) is highly dependent on the dataset used for training and evaluation. This paper presents a comprehensive comparative study using two distinct datasets: the benchmark Wireless Sensor Network Dataset (WSN-DS) and a custom-simulated dataset generated via OMNeT++ incorporating multiple routing protocols. We propose and evaluate a suite of models hyperparameter optimized with Particle Swarm Optimization (PSO), including PSO-MLP, against their standard counterparts (KNN, RF, NB, DT, and MLP). Our results demonstrate that while all models perform well on the generalized benchmark, the PSO-MLP model consistently achieves superior performance (up to 98.1% accuracy) on the complex, protocol-specific simulated dataset. This study highlights the critical importance of dataset selection and the significant performance gains achieved through metaheuristic optimization, providing a robust framework for developing intrusion detection systems (IDS) tailored to specific WSN architectures.