ADeepCRF: real-time threat detection in IoT network using blockchain-based advanced proof of authority and adaptive deep conditional random fields

An well-organized growth of smart services in the Internet of Things (IoT) communication system, the entire sensitive and crucial information is destroyed due to the occurrence of harmful attacks and behaviors in the data transmission performance. Therefore, the identification of threats in IoT system is significant for significantly protecting a large amount of sensitive data from hackers. In recent times, numerous threat detection frameworks have been designed for timely recognizing and localizing threat behaviors in IoT system. Yet, it is not suitable for easily recognizing subtle amounts of abnormal behaviors in an IoT system to make more inaccurate results in the detection process, minimizing earlier decision-making and generalization ability. Additionally, it requires more training duration and validation measures to successfully provide a better detection solution at dynamic IoT environmental conditions. In a communication system, blockchain technology was integrated with IoT to attain better data transmission. Executing threat detection in an IoT network through blockchain aids in controlling the tamperproof and decentralized nature of the network. Moreover, the blockchain technique secretly records and shares the essential information associated with the threats, which helps to identify the malicious activities over the network. In some cases, the auditabiity as well as transparency of the network are observed, and also the blockchain is considered as the immutable ledger to store the information. Furthermore, verifying the threats among several devices helps to recognize and react to cyber-attacks. So, this research work designs an intelligent machine model for real-time threat recognition in IoT utilizing blockchain technology. Here, blockchain-based Advanced Proof-of-Authority (BC-APoA) is introduced in this research work that utilize the potential of digital certificates rather than traditional data contracts, and it minimizes the overhead of the system, thus enhancing scalability and energy efficiency. Moreover, to tackle the dataset imbalance issue, an intelligent machine-learning model is used for threat detection. At first, the needed data are collected and given to the feature extraction process. Here, two sets of features are differentiated using a Support Vector Machine (SVM) and t-distributed Stochastic Neighbor Embedding (t-SNE). The obtained features are then given to the Adaptive Deep Conditional Random Fields (ADeepCRF) for further processing of the data. In this paper, we propose a novel Improved Random Variable-based Skill Optimization Algorithm (IRV-SOA) for timely tuning the significant, learning rate, steps per epoch and hidden neuron counts parameters of the ADeepCRF mechanism, enhancing better generalization and decision-making. Finally, the developed IRV-SOA model provides the threat-detected outcome. Then, the efficiency of the designed approach is proved by contrasted it with existing approaches. In the estimation phase, the proposed framework has achieved better accuracy values of 95.1%, 95.2%, 95.4%, 95.5% and 95.6% in terms of various hidden neuron counts, such as 100th, 200th, 300th, 400th, and 500th.