Optimized Bayesian regularization-back propagation neural network using data-driven intrusion detection system in Internet of Things

Nowadays, the network intrusion and cyberattack have emerged as the two main issues with Internet of Things (IoT) applications. The existing methods for preventing and detecting intrusions are limited in many ways, making it impossible to accurately identify any kind of attack occurring within network traffic. A number of machine learning-based methods that attains poor performance in the multiple class categorization accuracy are provided by the researchers. This research presents Data-Driven Intrusion Detection System in Internet of Things utilizing Optimized Bayesian Regularization-Back Propagation Neural Network (DIDS-BRBPNN-BBWOA-IoT) to overcome these issues. The input data is taken from TON_IoT Dataset. The data balancing of the training dataset is enhanced using Class decomposition and synthetic minority class oversampling method (CDSMOTE). Then, the data is pre-processed using Variational Bayesian-based Maximum Correntropy Cubature Kalman Filtering (VBMCCKF) of noise removal and data enhancement. The preprocessed output is given into feature extraction to extract features by using Dual-Tree Biquaternion Wavelet Transform (DTBWT). The extracted features are fed into Bayesian Regularization-Back Propagation Neural Network (BRBPNN) which detects the intrusion as Ransomware, Password attack, Scanning, Denial of Service (DoS), Distributed Denial of Service (DDoS), Data injection, Backdoor, Cross-Site Scripting (XSS), and Man-In-The-Middle (MITM). In general, BRBPNN does not show any optimization adaption methods to determine the optimal parameter for appropriate detection. Hence, Binary Black Widow Optimization Algorithm (BBWOA) is proposed in this manuscript to improve the BRBPNN classifier that detects intrusion precisely. The proposed DIDS-BRBPNN-BBWOA-IoT method is implemented using Python. The performance of the DIDS-BRBPNN-BBWOA-IoT approach is examined using performance metrics like accuracy, precision, recall, f1-score, specificity, error rate; computation time, and ROC. The proposed SAPVAEGAN-LCC-IR approach attains 18.44%, 26% ,and 29% greater accuracy; 26.55%, 24.12%, and 27.22% greater recall compared with existing MIDS-MIoT, AID-SDN-IoT, and IID-LW-IoT techniques.