A Dependable Hybrid Deep Learning Model for IoT Trust Management System

Trust management systems (TMS) are essential for securing internet of things (IoT) systems, particularly given the growing complexity and frequency of cyber-attacks. This research offers a hybrid model that combines two deep learning (DL) models to improve accuracy and identify untrustworthiness prediction, which is a challenge in managing large volumes of data, dynamic behaviours, and data sparsity. Our approach incorporates efficient pre-processing techniques and uses embedded techniques for behavioural feature selection. Subsequently, we use two models: Long Short-Term Memory (LSTM) and a hybrid Convolutional Neural Network (CNN) model. We evaluated these models using the transformed UNSW NB15 dataset. In our hybrid model, the role of CNNs is to identify spatial patterns in IoT systems and detect untrustworthiness, such as untrustworthy data patterns, which indicate potential threats, while LSTMs analyse these extracted behaviours over time, enabling them to recognise complex sequences and behavioural trends that develop into malicious activities. The LSTM model exhibited exceptional performance, with an accuracy of 98.8%, and a test accuracy of 98.7%. Furthermore, with a test accuracy of 99.5%, our hybrid deep learning-based TMS achieved the highest level of accuracy. Furthermore, we conducted a comparison between our results and other existing methods, demonstrating notable differences in performance when assessed using false alarm rate, detection rate, precision, F1 score, test accuracy, and test loss. By incorporating hybrid models, this study demonstrates a significant improvement in our TMS's performance in IoT systems.