Vampire Squid Optimization Algorithm for Bioinspired Energy Efficient Swarm Intelligence in Cyber Threat Detection

This paper presents the Vampire Squid Optimization Algorithm (VSOA), a novel energy-efficient, bio-inspired metaheuristic designed to optimize network-based intrusion detection systems (IDS) by jointly selecting features and tuning classifier hyperparameters. Inspired by the deep-sea vampire squid’s dual strategy of passive drifting and rapid striking, VSOA alternates between adaptive, diversity-driven exploration and shrinking-radius, Gaussian-based exploitation—executed under a strict evaluation budget to minimize computational overhead. VSOA is integrated with a Support Vector Machine (SVM) classifier using an RBF kernel, where each candidate solution encodes both a binary feature mask and real-valued hyperparameters (C, γ). Through iterative fitness evaluation, the algorithm searches for compact, high-performing IDS configurations that balance accuracy with efficiency. On the NSL-KDD dataset, VSOA achieves 95.0% accuracy, 93.0% recall, 94.0% precision, and a 5.0% false-positive rate, outperforming PSO and GA by 3.8% in accuracy and reducing false alarms by 22%. On UNSW-NB15, it reaches 93.2% accuracy and 92.5% F₁-score, again surpassing the best baseline by over 3.5%. VSOA converges rapidly reaching 90% accuracy in 60 ± 5 iterations (≈3,000 ± 200 evaluations), about half the time required by competing methods. It also reduces total evaluations and runtime by up to 25%, completing optimization in 120 seconds using 9,800 evaluations. These results confirm that VSOA, when coupled with SVM, delivers accurate, compact, and computationally efficient IDS solutions, making it ideal for real-time, resource-constrained environments such as IoT gateways and edge devices.