Cauchy Gaussian Mutation Strategy based Parrot Optimizer for Feature Selection in Intrusion Detection System

An expanding use of the Internet of Things (IoT) has heightened data security risks, making systems more vulnerable to attacks and resulting in breaches of critical data. Intrusion Detection System (IDS) is positioned in IoT networks for attacks detection and to ensure data security. However, the significant limitation of existing solutions is that they select flow-level features independently and avoid interactions among network flows, resulting in a degraded capability to detect threats effectively. Hence, to overcome these problems, this research proposes the Cauchy Gaussian Mutation Strategy based Parrot Optimizer (CGMS-PO) for the selection of relevant features in IDS. By selecting a minimal set of features without compromising performance, CGMS-PO minimizes the amount of data from the raw features, making the model more robust and efficient. Initially, the two standard datasets like ToN-IoT and Bot-IoT are utilized to experiment with the proposed method. The pre-processing techniques such as handling missing values, categorical features and Z-score normalization, are used to enhance data quality. The proposed CGMS-PO approach is used for feature selection and finally, Huber Loss function-based Support Vector Machine (HL-SVM) is taken out for the classification of network attacks into multiclass. The evaluation findings demonstrate that the proposed CGMS-PO approach achieves better accuracy of 99.99% and 99.01% on ToN-IoT and Bot-IoT datasets as compared to the stacked classifier and Ensemble Graph Convolutional Neural Network (GCN).