Maximizing Intrusion Detection in IoT Sector using Ensemble Network with Enhanced Heuristic-Assisted Weighted Feature Fusion Process
Abstract
Nowadays, the Internet of Things (IoT) sector is crucial for constructing smart paradigms. Privacy and security are taken as significant problems in any real-world applications dependent on the IoT system. The protection hurdles in IoT-aided devices create security threats that hinder smart sector development. Hence, there is an essential for Intrusion Detection Systems (IDSs) developed for IoT sectors to minimize the IoT-based security attacks that create several security issues. Conventional IDSs cannot be utilized in general IoT-aided networks because of the multiple attributes concerning computational capability, memory capacity, and battery life. To conquer these problems, multiple IDSs have been suggested in research works. But, various IDSs meet issues of low accuracy and high face alarm in the detection of anomalies. To address the vulnerabilities in the conventional systems, an ensemble deep learning mechanism is recommended to recognize the presence of intrusion in the IoT sector. Firstly, the raw data is collected from standard data sources. Furthermore, it is followed by feature extraction, which is done by utilizing Principal Component Analysis (PCA), T-distributed Stochastic Neighbor Embedding (TSNE) and statistical features. Subsequently, the overall features are fused and the weights optimally determined using the Fitness Value of Pelican Optimization Algorithm (FVPOA). Finally, the fused features are subjected to the Ensemble Networks (EnsNet), which are built with Long Short-Term Memory (LSTM), Deep Markov Random Field (Deep MRF), Recurrent Neural Network (RNN), and Ridge Classifier. Then, the final detected outcome is acquired by average score-based fusion. Therefore, the validation of the model takes place by employing divergent measures. On the contrary, the proposed work achieves a higher detection rate and also helps to evade third-party intrusion in the network.