Innovative detection of IoT cyber threats using a GBiTCN-Temformer and MKOA framework

Abstract

In the digital era, the Internet of Things (IoT) platform is progressively utilized in several applications due to its versatility, adaptability, and accessibility. Despite that, the lack of security protocols, improper device updates, and unauthorized access exposes the IoT environment to diverse cyber threats, which affect network security and user confidentiality. To enhance cybersecurity, various deep learning (DL) approaches have been developed, yet these techniques have struggled to detect various attacks affecting the IoT platform, often yielding imprecise results and high false positive rates. This research proposes a novel framework that leverages the Modified Kepler optimization algorithm (MKOA) based feature selection model and the Gated Bidirectional Temporal Convolutional Network (GBiTCN) based Temporal transformer (Temformer) detection model to improve the cybersecurity and integrity of the IoT devices. The MKOA technique is introduced to select optimal features from the normalized data, thereby analyzing the characteristics of attacks and reducing data dimensionality. In addition, this study introduces a hybrid detection model, which combines the Gated BiTCN and Temformer approaches. The Gated BiTCN captures the bidirectional semantic features of network traffic, while the Temformer extracts correlations among different sequences for precise cyber-attack detection. Further, the effectiveness of this proposed technique is measured by utilizing IoT attack datasets with distinct parameters and it attains a higher accuracy of 98.76 % during attack detection tasks. These results demonstrate that the proposed technique provides exceptional performance in cyber threat detection. Experimental validation shows that the model significantly contributes to cyber-attack detection and enhances the cybersecurity of IoT networks effectively.