An Optimized Multi-kernel Based Extreme Learning Machine for Authentication Threat Detection with Feature Reduction Scheme in IoT

Internet of Things (IoT) refers to a network of interconnected objects, computer systems, and physical or mechanical devices that are capable of sending and receiving data over a network without the intervention of a human. A growing number of undesirable security issues are occurring as a result of the rapid advancements in IoT infrastructures. IoT devices are therefore becoming more vulnerable to security threats and network attacks. IoT services and innovative ecosystem devices can be severely damaged as a result of these attacks and issues. So we propose a multi-kernel-based intensive learning machine model to improve security and overcome the above challenges. To enhance the performance of the insider threat detection model, significant feature sets are generated using correlation coefficient, random forest mean (RFM) reduction accuracy, and gain ratio. To obtain an optimal feature set, the features are combined using an appropriate mechanism (AND function). Afterward, the combined feature set is fed into a machine-learning model based on multi-kernel optimization (MKELM). MKELM is optimized using the Adaptive Crocodile Optimization Algorithm (ACOA) to increase its performance in this detection model. Java is used for the implementation of this proposed approach. In order to analyze performance, NSL-KDD 99 dataset and UNSW-NB15 dataset have been used. The performance of the proposed scheme is evaluated in terms of detection rate, precision, accuracy, recall, F-measure, and detection rate.