Next-generation industrial networks incorporate diverse devices and technologies, including cloud- and fog-based systems and large-scale Industrial Internet of Things (IIoT) applications. These heterogeneous environments comprise thousands of sensors, actuators, controllers, and geographically distributed zones, posing significant challenges for authentication mechanisms that must scale while remaining secure and practical under real industrial conditions. Current blockchain-based decentralized identity-management systems are characterized by high transaction overhead, reliance on computationally intensive cryptographic operations, and partial dependence on trusted entities, all of which limit their scalability and practical deployment. To address these issues, in this paper, we propose a lightweight, blockchain-based identity-management solution that integrates epoch- and batch-based registration with sparse Merkle trees (SMTs). Device registrations are grouped into fixed epochs, and cryptographic costs are amortized across multiple devices to substantially reduce per-device overhead. The SMT structure generates verifiable, reusable proofs that preserve privacy while enabling efficient verification; sensitive device information is never revealed during authentication. Fog nodes act as decentralized verifiers, supporting distributed trust without continuous interaction with the blockchain. Experimental results show that the system can be practically deployed on Raspberry Pi and ESP32 devices, reliably perform identification with around 1000 total messages, and operate at the tested upper bound of 7000 devices distributed over 20 fog nodes, all while keeping the blockchain storage overhead under 5 MB.

This research presents a robust framework for the multi-classification of URLs into benign and malicious categories, specifically addressing defacement, malware, phishing, and spam. We aggregated multiple publicly available datasets into a unified, balanced dataset, followed by rigorous preprocessing to enhance data quality. Structural URL features-including length, subdomain count, HTTPS presence, digit and special character ratios-were systematically extracted. The model uses DistilBERT transformer embeddings to turn the URLs into semantic representations. Experiments are done using XGBoost, LSTM, and fine-tuned DistilBERT. Results show that the DistilBERT model has an accuracy of $97.4 \%$, which is better than traditional ensemble methods and recurrent neural networks. Merging the structure of data with its meaning gave better results, proving that transformer-based approaches are valuable for cybersecurity.

ABSTRACT The proliferation of the Internet of Things (IoT) has reshaped industries based on seamless connectivity. However, it has also brought about immense security challenges, especially in the communication protocol of routing protocol for low‐power and lossy networks (RPL). One of these security threats vital to the RPL‐based IoT networks includes the Clone ID attack on malicious nodes when they clone the identity of legitimate nodes to access their sensitive data without authorization. Detecting Clone ID attacks in RPL‐based IoT networks is complex because network traffic data has high dimensions and substantial data imbalances while facing limited resources in these environments. The unmanaged control message system and insufficient identity authentication methods within the RPL protocol directly expose networks to state‐of‐the‐art cyber security threats. This paper proposes a new edge layer‐based deep neural network (DNN) approach to detect Clone ID attacks from IoT sensor networks by network traffic pattern analysis. The proposed method is based on deep data features to distinguish legitimate nodes from cloned nodes and improve the overall security, resilience, and operational efficiency of RPL‐based IoT networks. To check the efficiency of our proposed method, we designed a synthetic dataset called CID‐RPL. The CID‐RPL dataset consists of 25 attributes and 2,131,328 samples. The experimental results are best to describe that our proposed approach outperformed the previously designed methods by offering an accuracy improvement of 5.06%, precision improvement of 7.60%, recall increment of 7.0%, and F1 score enhancement of 11.0%. Similarly, residual energy at the network level increased by 32.84%, which infers that the lifetime of the network will be extended and its energy efficiency increased under attack situations. Thus, the results testify to the effectiveness of the DL‐based solution proposed herein to detect Clone ID attacks in dynamic and evolving network environments.

Next-generation industrial networks are designed to integrate a wide range of devices, services, and applications spanning multiple technologies, such as cloud platforms, edge computing, and the Internet of Things. With the growing adoption of applications such as “Industry 4.0,” high security and low latency are becoming unavoidable requirements for these networks. Traditional virtual private networks (VPNs) generally experience performance, latency, and security issues, especially when supporting secure remote access for Industry 4.0 and ehealth applications. To address these issues, this study introduces a novel zero-trust network-access framework for next-generation industrial networks called Secure Transmission (SecT). SecT is a User Datagram Protocol (UDP)-based solution, ensuring speed and effectiveness, with role-based access control. It uses a centralized management interface that can adapt to various network environments, providing secure access to mission-critical applications and increasing operational agility. SecT aims to meet the emerging demands of modern industrial networks, offering secure access with improved performance. The results of a comparative analysis show that SecT outperforms traditional VPNs in both capability and flexibility, adapting well to new network conditions.

Summary The rapid growth of the Internet of Things (IoT) has led to its widespread adoption in various industries, enabling enhanced productivity and efficient services. Integrating IoT systems with existing enterprise application systems has become common practice. However, this integration necessitates reevaluating and reworking current Enterprise Architecture (EA) models and Expert Systems (ES) to accommodate IoT and cloud technologies. Enterprises must adopt a multifaceted view and automate various aspects, including operations, data management, and technology infrastructure. Machine Learning (ML) is a powerful IoT and smart automation tool within EA. Despite its potential, a need for dedicated work focuses on ML applications for IoT services and systems. With IoT being a significant field, analyzing IoT‐generated data and IoT‐based networks is crucial. Many studies have explored how ML can solve specific IoT‐related challenges. These mutually reinforcing technologies allow IoT applications to leverage sensor data for ML model improvement, leading to enhanced IoT operations and practices. Furthermore, ML techniques empower IoT systems with knowledge and enable suspicious activity detection in smart systems and objects. This survey paper conducts a comprehensive study on the role of ML in IoT applications, particularly in the domains of automation and security. It provides an in‐depth analysis of the state‐of‐the‐art ML approaches within the context of IoT, highlighting their contributions, challenges, and potential applications.

In recent years, the Internet has facilitated the emergence of social media platforms as significant channels for individuals to express their thoughts and engage in instantaneous interactions. However, the reliance on online reviews has also given rise to deceptive practices, where anonymous spammers generate fake reviews to manipulate the perception of a product. Ensuring the integrity of the online review system requires identifying and mitigating fake reviews. While existing machine learning (ML)- and neural network (NN)-based sentiment analysis methods can detect deceptive reviews, they often suffer from long training times, high computational resource requirements, and memory constraints. This study aims to overcome these limitations by introducing a transformer-based “deceptive bidirectional encoder representations from transformers (DeceptiveBERT) model.” This model utilizes contextual representations to enhance the precision of deceptive review identification. Transfer learning is employed to leverage knowledge from a pre-existing BERT base-uncased word embedding model, enabling efficient feature extraction. The proposed model incorporates a combination of classification layers to categorize reviews into two distinct categories: deceptive and truthful. Additionally, the study addresses the challenge of imbalanced datasets by utilizing three separate datasets and implementing appropriate methodologies for dataset curation. The effectiveness of the DeceptiveBERT model was evaluated through experimentation. The results demonstrate its efficacy, with the model achieving accuracy rates of 75%, 84.79%, and 81.08% on the Ott, YelpNYC, and YelpZip datasets, respectively.

Data publication and sharing are critical components of assessing network infrastructures in the Internet of Things for quality-of-service enhancement. Especially, the advancement in communication technology (e.g., 5G/6G) enables the improvement of the current bottlenecks in the Industrial Internet of Things. Recent approaches remove raw data and their source to achieve a privacy guarantee. However, the data are already anonymized; these still reveal the victim’s extra information using linkage attacks. When data are updated and combined or noise is introduced as part of conventional privacy protection approaches, such as$k$-anonymity, l-diversity, or differential privacy, the usefulness of the released data is diminished, however, posing data utility and computation constraints. In recent years, lightweight privacy-preservation techniques have been proposed for these reasons. However, most of the focus is on syntactic privacy instead of semantic privacy guarantee. Therefore, this article proposes a lightweight semantic privacy-preservation framework for maintaining privacy with high utility efficiency. The proposed paradigm ensures semantic privacy by combining probabilistic random sampling with Instant_Anonymity. Compared to$k$-anonymity, the suggested model demonstrates improved data utility with lower utility errors of 0.00036 and 0.41 for Kullback–Leibler divergence and query error, respectively. The classification accuracy is improved by 0.2%. In addition, the proposed approach is simpler to implement in computation time than the existing state-of-the-art lightweight privacy-preserving strategies.