Optimizing anomaly detection models for edge IIoT with an enhanced firefly algorithm-based hyperparameter tuning strategy

Abstract

• An improved firefly algorithm (EFA) is used to optimize hyper- parameters in a novel machine learning (ML)-based NIDS framework for Edge-IIoT scenarios. • The effect of different EFA control settings and other meta-heuristic approaches on intrusion detection machine learning models is examined. • The convergence of fitness values in control parameter optimization is used to assess the efficacy of meta-heuristic algorithms. • The advantage of EFA-optimized models over alternative metaheuristic algorithms is confirmed by statistical testing (Friedman and Wilcoxon's tests). • Accuracy, precision, recall, and F1-score are all improved when the suggested EFA-optimized ML models are thoroughly compared to current methods. Security issues in the Industrial Internet of Things (IIoT) have grown more serious as industrial automation rises as these networks are especially prone to cyberattacks. By means of adaptive attack detection models, machine learning (ML) presents a potential solution. High-dimensional hyperparameter spaces, insufficient labeled data for different attack labels, limited computational resources on edge devices, and the need for real-time detection are just a few of the difficulties classical ML techniques confront, though. This work presents an Enhanced Firefly Algorithm (EFA) for optimizing the hyperparameters of ML models to enable effective and accurate threat detection in resource-constrained IIoT contexts, so addressing these challenges. Reliable cyberattack detection is the main goal; hence, minimizing computing overhead during deployment is crucial for both real-time and edge-based solutions. The suggested EFA enhances convergence toward the global optimum by including a suitable stochastic component generator that improves both exploration and exploitation ability. Furthermore, the approach uses Synthetic Minority Oversampling Technique (SMOTE) to handle the class imbalance in the Edge-IIoT dataset. To show the superiority of the proposed method, several ML classifiers are trained using EFA optimized hyperparameters and compared against models customized via various conventional metaheuristic techniques. Along with training and testing schedules and statistical significance tests (Friedman and Wilcoxon), performance is assessed using conventional measures like accuracy, precision, recall, and F1-score. The EFA offers improved resilience and efficiency since the results reveal that it constantly ranks the best average. These results support the efficiency of the suggested approach for IIoT cybersecurity applications in real-time.

Topics & Concepts