Cognitive Cybersecurity via Bio-Inspired Intelligence: A Starling Flock Optimized Spiking Neural Network for Resilient Mobile Threat Detection

Abstract

Real-time threat detection is critical for guaranteeing mobility systems in cognitive environments to function with little disruption. Through monitoring user behavior, system logs, and network traffic, such systems depend on sophisticated analytics and machine learning. This paper presents a new Intrusion Detection System (IDS) architecture leveraging a Spiking Neural Network (SNN) as a classifier and the Starling Flock Optimization Algorithm (SFOA) as a feature selector on the NSL-KDD dataset. The main objective is to enhance the accuracy of classification while lowering feature dimensionality and computational cost for effective real-time detection. SFOA, stimulated by starling flock behavior, dynamically chooses best feature subsets with balance between relevance and redundancy. The SNN, in turn, simulates biological neurons and processes these best features by mapping them into temporal spike trains, employing event-driven processing and Spike-Timing-Dependent Plasticity (STDP) for detailed pattern detection. Our developed model shows excellent performance with accuracy of 99.02%, precision of 98.9%, recall of 99.18%, and a 99.04% F1-score, outperforming SVM, Random Forest, CNN, LSTM, and Transformer models. Based on only 18 chosen features, the model has an extremely fast inference time of 2.3 ms, which is perfect for real-time usage. Comparatives also confirm SFOA is superior to PSO, GWO, ACO, and DE in terms of convergence rate and feature compactness. Ablation experiments validate that SFOA, spike encoding, and STDP learning are all essential to the model's performance, making this hybrid architecture a scalable and energy-efficient solution for sophisticated intrusion detection.