Enhanced stacked autoencoder with hybrid steller model for intrusion detection with Chebyshev artificial protozoa remora algorithm
Year: 2026
Type: article
Abstract: The Internet of Things (IoT) networks intrusion detection is still difficult due to dynamic attack trends, high false positives, the volume of data, and the possibility to detect temporal relationships in network traffic. The current optimization-based IDS models have been known to face the challenge of balancing between exploration and exploitation in their approach to feature selection, which results in sub-optimal detection performance. Furthermore, they typically fail to achieve a compromise between robust detection accuracy and computing efficiency. This paper suggests a novel hybrid deep learning-based intrusion detection optimization system to address these problems. To put the data in the standard form, the input from the AWID dataset is first pre-processed using Min-Max normalization. Using the normalized data, the Enhanced Remora Optimization (EReO) algorithm is employed to select the significant attributes for improving the accuracy and efficiency of intrusion detection. Here, EReO integrates Chebyshev mapping and artificial protozoa optimization (APO) algorithm within the traditional Remora Optimization Algorithm to improve the exploration and eliminate the issue of local optimal trapping. After this, the proposed Enhanced Stacked Auto Encoder Based Simple Recurrent Unit (ESAE_SRU) model, which incorporates a hybrid stellar-based cleaner fish optimization (HSCFO) algorithm, performs the intrusion detection task. This model is tested on the AWID data with k-fold cross-validation (k = 2–10), and its performance is measured in terms of accuracy, precision, and recall, F1-score, false positives, and Matthews correlation coefficient (MCC). The proposed model attained 99.86% accuracy, 99.56% recall, 99.82% precision, 99.66% F1-score, 0.108% FPR, and 99.99% MCC at 10-fold cross-validation, which is better than the existing methods.
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,880
OpenAlex ID: W7215894855