Novel Methods Applied to Security and Privacy Problems in Future Networking Technologies

1. Introduction

The rapid development of future networking technologies, such as 5G, 6G, blockchain, the Internet of Things (IoT), cloud computing, and Software-Defined Networking (SDN) is set to revolutionize our methods of connection, communication, and data sharing. While these technologies facilitate more efficient communication and enable the creation of previously unimaginable applications and services, they also pose significant security and privacy challenges. These include expanded attack surfaces, new attack vectors, evolving threats, and concerns over data privacy and user trust. Addressing these challenges is crucial to ensuring that future networking technologies can be utilized safely and securely while realizing their full potential.

This Special Issue is particularly timely, as it highlights cutting-edge research and solutions aimed at overcoming these security and privacy challenges. Recent advancements in networking technologies and security frameworks have laid a strong foundation for tackling long-standing issues in secure communication and data privacy. By showcasing both theoretical innovations and practical applications, the Special Issue aims to inspire further advancements in the field. The objective is to bridge the gap between theory and practice, promoting collaboration across various disciplines to develop more resilient and secure networking infrastructures that can address the emerging security and privacy concerns of our digital future.

In this Special Issue on cybersecurity, we received 19 submissions, each carefully evaluated by at least one of the Guest Editors to ensure relevance to the theme of securing emerging technologies. Submissions considered relevant underwent a comprehensive review process involving at least two external reviewers, while those that did not meet the necessary criteria were rejected. After a rigorous peer-review process, 11 articles were selected for publication. These contributions explore critical topics such as the development of robust cybersecurity strategies for emerging technologies, including 5G, blockchain, IoT, cloud computing, and SDN, as well as Zero Trust frameworks. The articles provide innovative insights into how these technologies can be safeguarded against evolving security threats, addressing issues such as expanded attack surfaces, data privacy concerns, label-flipping attacks, and user trust issues. Key technologies such as Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), Deep Reinforcement Learning (DRL), and blockchain for identity management, along with Zero Trust principles for enhancing security, are applied to tackle cybersecurity challenges. A summary of the findings and conclusions from each article is presented below, offering a diverse range of solutions for securing these rapidly advancing technologies in the face of complex cyber threats.

2. Review of Published Papers

In the first contribution, Zhang, Wu, and Wang present a novel generic algorithm (DE-C&W) for adversarial example generation to enhance deep learning security evaluations. The study aims to improve upon existing methods by reducing computational costs and increasing the transferability of generated examples and query times. The authors combine the Differential Evolution (DE) algorithm to preprocess images, identifying vulnerable pixels and reducing the dimensionality of the search space, with the Adam optimizer to refine the C&W attack, resulting in a more efficient and effective method for generating adversarial examples. Their findings demonstrate the DE-C&W algorithm’s superior performance in attack success rate, transferability, and reduced query numbers and attack costs, positioning it as a viable solution for stress-testing the robustness of deep learning models.

In the second contribution, Mutar et al. propose a Markov-based predictive model called Markov Detection and Prediction (MDP) for mitigating Distributed Denial of Service (DDoS) attacks in Long-Term Evolution for Machine (LTE-M) networks within the Internet of Things (IoT). The study introduces a Continuous-Time Markov Chain (CTMC) model to predict and classify network traffic into three categories: Authentic, Suspicious, and Malicious, leveraging the MDP model to preemptively identify potential DDoS threats by analyzing system behavior under varying traffic conditions. Their findings, demonstrated through extensive simulation results, showcase the model’s ability to successfully forecast the impact of DDoS attacks and effectively differentiate between traffic types, allowing for timely intervention in Machine-to-Machine (M2M) communication environments and highlighting its potential for enhancing the security and reliability of IoT systems against evolving cyber threats.

In the third contribution, Ashawa et al. present an enhanced image-based malware classification system that uses Convolutional Neural Networks (CNNs) with ResNet-152 and Vision Transformer (ViT) architectures to improve malware detection, addressing the challenge of accurately classifying complex malware variants. The authors convert executable files into grayscale images and analyze content distribution features and pixel intensities to distinguish between benign and malicious software. The proposed system achieves a high-performance accuracy of 99.62% using 10-fold cross-validation with efficient computation. It makes it suitable for dynamic and complex malware environments and highlights the importance of feature normalization and dimensionality reduction. Experimental results indicate that the model effectively combines structured and content-based malicious features and can differentiate between benign and malicious software, making it a viable solution for real-time malware detection.

In the fourth contribution, Chang et al. propose a novel coverless data hiding method that leverages texture images generated from Two-Dimensional Fractional Brownian Motion (2D FBM) and deep learning models to address the limitations of traditional techniques relying on common, meaningful images and their vulnerability to detection. The approach focuses on generating texture-like cover images, known as Fractional Brownian Images (FBIs), for encoding secret messages by adjusting the Hurst exponent of the 2D FBM process, and utilizes deep learning models as a key tool to select qualified FBIs from a generated database to form stego images, creating a mapping between these code images and secret data, enhancing security. Experimental results demonstrate that the proposed scheme can adaptively hide data in a variety of natural textures, such as clouds or marble patterns, offering high embedding capacity and robust security against steganalysis and transmission attacks, making it a flexible, reliable, and high-capacity method particularly viable for scenarios with limited secret data.

In the fifth contribution, El-Hajj examines the integration of Digital Twins (DTs) with Intrusion Detection Systems (IDSs) to enhance the cybersecurity of IoT-based smart city infrastructures by providing real-time threat detection and performance monitoring. The study proposes a solution that utilizes Eclipse Ditto as the DT platform for managing Digital Twins and Snort as the IDS platform, creating a testbed on a Raspberry Pi device to simulate common cyberattacks such as Hping3 flood attacks and NMAP reconnaissance scans. The findings indicate that this combination allows for real-time monitoring, provides a dual perspective on system security and performance, offers new insights into vulnerability identification (noting IDS effectiveness against flood attacks but struggles with NMAP scans, suggesting optimization needs), and highlights the impact of these attacks on system resources like high CPU and memory usage, demonstrating significant potential for bolstering the security and resilience of smart city environments by leveraging Digital Twins for both protection and simulation.

In the sixth contribution, Iffländer et al. propose a dynamic, attack-aware framework for Security Service Function Chain (SSFC) reordering to optimize the performance and improve the security of network infrastructures, addressing the limitations of traditional static SSFCs that do not adapt to the type of attack or traffic observed. The authors introduce the concept of dynamically reordering security functions within an SSFC based on real-time attack detection, which reduces resource demands and improves overall security efficiency, demonstrating through performance modeling and experimental validation that reordering security functions based on the type of attack can lead to up to a 59% reduction in required computational resources while highlighting that no single static order outperforms others across all attack types. Their proof-of-concept framework, which uses a Function-Chaining Controller (FCC), allows for this dynamic reordering and can be implemented within SDN-enabled networks, helping mitigate resource bottlenecks and enhance the scalability of security infrastructures in real-time environments, indicating the potential for more resilient and efficient cybersecurity systems.

In the seventh contribution, Neto et al. introduce the Overlapping Label Recovery (OLR) framework to counter the effects of label-flipping attacks that degrade Deep Learning-based Internet of Things (IoT) profiling by utilizing Random Forests (RF) as internal cleaners to recover corrupted labels and re-evaluating the dataset to minimize the impact of these data poisoning attacks, which manipulate training labels to degrade model performance. The framework employs overlapping training samples, and the authors validate the approach using a poisoned version of the CIC IoT Dataset 2022, demonstrating that the OLR strategy consistently improves IoT profiling performance across various metrics, including accuracy, recall, precision, and F1-score, while highlighting the importance of controlling the training overlap to maintain high performance and suggesting that OLR can significantly mitigate label-flipping attacks in IoT environments.

In the eighth contribution, Daah et al. propose an enhanced Zero Trust (ZT) framework specifically designed for the financial industry by integrating blockchain technology to improve cybersecurity, acknowledging the limitations of traditional security models and the increasing sophistication of cyber threats. The proposed framework improves traditional ZT paradigms by introducing blockchain to enhance identity and access management (IAM), device and network security, and data protection, ensuring tamper-proof verification and enhanced data integrity. The study adopts a mixed-methods approach, combining qualitative analysis (literature review, threat modelling) with developing a prototype banking application for vulnerability scanning and performance testing to improve security while maintaining operational efficiency. Experimental results demonstrate that the framework offers a comprehensive security solution, fostering greater consumer trust while efficiently mitigating risks like advanced persistent threats and enhancing security and scalability.

In the ninth contribution, Ho-Oh et al. explore the use of Deep Reinforcement Learning (DRL) to simulate cyber-attacks and enhance cybersecurity measures, acknowledging the limitations of traditional defense mechanisms against evolving threats and leveraging DRL algorithms such as Deep Q-Network (DQN), Actor-Critic, and Proximal Policy Optimization (PPO) to develop adaptive solutions. The study adapts these DRL algorithms to simulate cyberattacks in a controlled environment. It uses the MITRE ATT&CK framework to model realistic attack scenarios and custom reward structures, adversarial training, and dynamic environments, comparing their effectiveness with the traditional Q-learning algorithm. The results demonstrate that the Actor-Critic algorithm outperforms the others in terms of success rate, learning efficiency, and reward acquisition, requiring the fewest iterations to complete each episode and showcasing its superior adaptability in dynamic environments, highlighting the potential of DRL in developing more intelligent and adaptive cybersecurity systems capable of evolving with increasingly sophisticated cyber threats and creating resilient cybersecurity systems.

In the tenth contribution, Agrawal et al. examine the application of Generative Adversarial Networks (GANs) for generating synthetic attack data to support and enhance cybersecurity efforts, addressing the challenge of acquiring realistic cyberattack datasets due to privacy concerns and the difficulty of obtaining large, diverse data for training deep learning models. The paper highlights GANs’ ability to generate high-quality synthetic data spanning diverse domains. It explores their potential to create realistic cyberattack datasets that can be used to train machine learning-based intrusion detection systems, scrutinizing data generation capabilities and assessing the value of synthetic attack data in training intrusion detection classifiers capable of detecting new and unseen real-world attacks, while including the most recent techniques by the scientific community and highlighting the key benefits and challenges of using GANs in cybersecurity, particularly in generating diverse and accurate synthetic attack data for robust model training and anomaly detection.

In the final contribution, Abdulhamid et al. investigate the current state of safety and security analysis frameworks for the Internet of Things (IoT), providing an overview and addressing the need for reliable and risk-free IoT applications by analyzing both classical and Model-Based Systems Engineering (MBSE) approaches. The survey reveals that most analysis frameworks are based on classical manual approaches, which tend to be time-consuming and error-prone, and only partially address cyber-security issues or the complex interactions and interdependencies between safety and security, highlighting the limitations of existing MBSE approaches that are still in their early stages and have not adequately addressed these interactions. The authors conclude by proposing a research direction for developing a novel MBSE approach for a unified treatment of safety and security requirements in the IoT domain, calling for the development of a unified approach to coanalyze safety and security properties and suggesting future research directions include exploring new modeling techniques and software-based analysis frameworks to assess better the safety and security of dynamic and evolving IoT environments.