An Effective Clustering and CSRvNN Model for Intelligent Intrusion Detection for the Internet of Things

Abstract

The identification of network intrusions is the primary function of an Intelligent Intrusion Detection System (IIDS), and it is essential to maintaining the protection of the Internet of Things (IoT). Deep learning (DL) was quite successful in identifying intrusions. Although the actual deployment of DL-based high-complexity frameworks is hampered by the limited processing power and data storage of IoT equipment. Moreover, DL-based anomaly identification is frequently linked to high false alarms with reduce precision and detection rates if it is struggling to precisely identify a variety of threats. This work proposes a hybrid learning method through combination of Adaptive K-Means (AKM) clustering and Context Sensitive Recursive Neural Network (CSRvNN) for the recognition and sorting of IoT threads. The suggested method entails grouping all of the information into a relevant cluster before utilizing a classifier to do categorization. The training instances are initially divided into clusters utilizing the Manhattan distance and AKM. The network dataset's dimensionality was decreased, and in order to enhance the suggested approach, pertinent features were chosen from the dataset using the Assimilated Artificial Fish Swarm Optimization (AAFSO) method. The generated clusters, which show a density zone of normal or abnormality cases, used as the foundation for an FFO-CRvNN classification algorithm. This is useful in assessing how well the clustering finds hidden attack patterns in the data.  The outcomes of the experiments demonstrate that the suggested model is more successful than the conventional categorization method in identifying assault patterns.