Adaptive Intrusion Detection in MANET Environments using PDF-Enhanced Quantized Spiking Neural Networks

Mobile Ad-hoc Networks (MANETs), characterized by their dynamic, decentralized, and infrastructure-less nature, are highly susceptible to security threats such as Black Hole (BHA), Wormhole (WHA), and Grey Hole (GHA) attacks. The inherent limitations of MANETs, including restricted bandwidth and device heterogeneity, often compromise the accuracy of traditional Intrusion Detection Systems (IDS). To address this, we propose a novel technique: the Probability Density Function-Quantization Spiking Neural Network (PDFQSNN), specifically tailored for the MANET environment. The PDF component is integrated into the neural network to prioritize data instances that are statistically more likely to represent significant network behaviors or attack patterns, thereby improving classification robustness. Furthermore, the Adaptive Moth Flame Optimization (AMFO) algorithm is employed for feature selection, effectively reducing the high dimensionality of the simulated dataset and isolating the most critical features for intrusion detection. Tested on a simulated dataset encompassing BHA and WHA scenarios, the PDF-QSNN method demonstrates superior performance compared to existing deep learning models like Convolutional Neural Networks (CNN) and Stacked Recurrent Long Short Term Memory (SRLSTM). Specifically, the proposed model achieves a high classification accuracy of 92.86% for BHA and 91.45% for WHA, validating its efficacy as an adaptive and accurate IDS solution for dynamic MANETs.