Optimized Hybrid Fuzzy NFSC Approach for Enhanced Scalable and Efficient Anomaly Detection in Secure IoT- Systems with Low-Latency
نویسندگان
1 Rajkot
2 Department of Computer Science and Engineering, Vidarbha Institute of Technology, Nagpur, India.
3 Department of Computer Engineering, Marwadi University, Rajkot. Gujarat, India
4 Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Andhra Pradesh – 522302, India.
doi
10.22111/ijfs.2025.51473.9092چکیده
The rapid growth and widespread integration of the Internet of Things (IoT) across various domains have enabled systems to autonomously send, receive, and analyze data. Despite its significant benefits for improving quality of life and service efficiency, IoT still faces considerable challenges related to security and privacy. An anomaly-based Intrusion Detection System (IDS) can provide security for IoT networks against various cyber-attacks. This study proposes an anomaly-based IDS to prevent cyber-attacks in IoT environments. In this research, we propose a novel optimized ensemble deep learning-based approach for anomaly-based IDS classification. The proposed approach includes several innovative aspects: initially, raw data are pre-processed, and essential features are extracted using CoffeeNet. The approach employs the Binary Bamboo Forest Growth Optimization Algorithm (BBFGOA) to reduce data characteristics and enhance anomaly detection effectiveness. We use the Optimized Fuzzy NFSC approach an ensemble of the Sharpened Cosine Similarity-based Neural Network (SCS-Net) and a rule-based Deep Fuzzy System with Nonlinear Fuzzy Feature Transform (NFFS-DFS)—to categorize network flows as benign or malicious across multiple categories. The Partial Reinforcement Optimization Algorithm (PRO) is used to tune the hyperparameters in the Fuzzy NFSC approach. Additionally, we employ a Conditional Tabular Generative Adversarial Network (CTGAN) to address the class imbalance issue. The proposed anomaly detection approach is evaluated and analyzed using four benchmark datasets to validate its effectiveness. The evaluation's results demonstrate that the approach enhances detection performance, with a 99.43% F1-score, 99.61% accuracy, 99.45% precision, and 99.38% recall. The findings also show that employing an oversampling technique improves minority sample identification rates, which raises overall accuracy. The suggested approach works better than current techniques since it takes less time to train and detect.
