Work

FGWO: A Non-Markovian Long-Memory Grey Wolf Optimizer for MLP Hyperparameter Tuning in Intrusion Detection Systems

Year: 2026

Type: conference-paper

Abstract: Grey Wolf Optimizer (GWO) may prematurely converge due to its memoryless update rule. We propose FGWO, a non-Markovian enhancement that injects fractional-order-inspired long-term memory via power-law weighting of historical states while maintaining GWO’s low computational cost. FGWO integrates fitness-guided Levy flights and an adaptive local refinement stage to balance exploration and exploitation. On the CEC2017 benchmark suite, FGWO consistently outperforms standard metaheuristics in terms of Success Rate (SR) and Mean Function Evaluations to Success (MFES). In a practical case study, FGWO tunes multilayer perceptron hyperparameters for intrusion detection on UNSW-NB15, achieving a 98.03% F1-score under a fixed evaluation budget. These results suggest that fractional-inspired long-term memory provides an effective and lightweight enhancement to GWO for both benchmark and cybersecurity optimization tasks.

Source: International Wireless Communications and Mobile Computing Conference (IWCMC)

DOI: 10.1109/iwcmc69287.2026.11579815

Authors: Nasreddine Hamdi, Akram Belazi, Safya Belghith, Héctor Migallón

Institutions: National Engineering School of Tunis, Tunis El Manar University, Universitat de Miguel Hernández d'Elx

Language: English

Topic: Metaheuristic Optimization Algorithms Research

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

OpenAlex ID: W7167285372

FWCI:

Cites: 18

Cited by:

Related to: 10