Modeling of Bayesian-Based Optimized Transfer Learning Model for Cyber-Attack Detection in Internet of Things Assisted Resource Constrained Systems

Security donates itself as one of the largest attacks on the support and development of the Internet of Things (IoT). Security challenges are evident in cyber-security threads that direct the main Internet service provider and weaken a significant part of the complete Internet by benefiting from defective and vulnerable embedded gadgets. Numerous devices inhabit at-home systems with user-administrators unfamiliar with network security best practices, creating simple goals for the attackers. So, security solutions are required to direct the untrusted and insecure public networks by mechanizing defences over affordable and nearby direct network data sharing. The growth of automatic cyberattack classification and detection tools utilizing artificial intelligence (AI) and machine learning (ML) devices become vital to achieving safety in the IoT environment. It is desired that safety issues allied to IoT devices be effectively diminished. This article proposes an Advanced Ensemble Transfer Learning for Cyberthreat Detection in Low Power Systems (AETL-CDLPS) technique. The primary intention of the AETL-CDLPS technique is to automate the detection of cyber-attacks for IoT-assisted resource-constrained systems. The AETL-CDLPS technique utilizes a linear scaling normalization (LSN) model to normalize the input data. Next, the AETL-CDLPS technique employs an improved coati optimization algorithm (ICOA)-based feature selection technique to choose optimal features. For the cyber threat detection process, an ensemble transfer learning (TL) model comprises three classifiers, namely gated recurrent Unit (GRU), deep convolutional neural network (DCNN), and stacked sparse autoencoder (SSAE). Finally, the Bayesian optimization algorithm (BOA) is utilized to optimize the hyperparameter tuning of the three ensemble techniques. The AETL-CDLPS model’s performance validation is performed using the Bot-IoT dataset. The comparison study of the AETL-CDLPS method portrayed superiorAccuracy, Precision, Recall, and F-Score values of 99.19%, 96.10%, 95.97%, and 96.03% over existing models.