Abstract Deploying unsupervised anomaly detection systems in heterogeneous smart home environments is hindered by the need for costly, per-site hyperparameter tuning. This paper addresses the critical challenge of hyperparameter transferability for creating zero-tune, plug-and-play security solutions. We systematically evaluate five unsupervised machine learning models [Elliptic Envelope (EE), Isolation Forest (IF), Local Outlier Factor (LoF), One-Class SVM (oSVM), and an Autoencoder (AE)] across five prominent IoT datasets. Using a rigorous dataset-specific hyperparameter tuning approach, we benchmark the performance of transferred configurations against both per-dataset optimization and default settings. Our findings establish a clear performance hierarchy: while dataset-specific tuning remains the gold standard, an intelligent transfer strategy significantly outperforms default configurations. Notably, we identify the IoTID20 dataset as the most effective source. Our quantitative topological analysis supports this, revealing that IoTID20’s high feature space complexity and cluster overlap (evidenced by low Silhouette scores) create a rigorous training environment that produces robust, portable hyperparameters. Furthermore, our analysis reveals a strategic trade-off: Autoencoders and LoF deliver the highest absolute performance, whereas IF offers the most substantial improvement over default settings. This work provides a quantitative framework for dataset-driven initialization, guiding the development of robust, low-maintenance intrusion detection systems.

The prospect of ultra-dynamic and tailored network slicing in the era of Beyond $\mathbf{5 G}$ (B5G) is associated with the potential to make networks more vulnerable to stealthy adversarial actions. Attackers may impersonate authorised end users, saturate virtual slices, or take advantage of resource gaps at the expense of service availability, quality and user trust. This study presents a novel approach, Secure Intelligent Enforcement of Guaranteed Efficiency (SIEGE), which is a strategic defence of scarce resources that are under siege by optimising resource allocation with adversarial awareness in the core of the defence. SIEGE is created through the use of an Integer Linear Programming (ILP) and is capable of identifying and blocking malicious usage by users without impacting the performance of benign users. The model proposes per-user behavioural metrics ($p_{u}, q_{u}$), which measure slice/resource exposure, and applies them in an optimisation target that is resilience-aware. Early indicators suggest a maximum of $43.7 \%$ reduction in resource hijacking cases, while service-level agreements remain intact. The research is continuing to provide the foundation to a next-generation secure slicing paradigm, which is proactive, explainable and scalable. SIEGE provides a roadmap of security-by-design wireless infrastructure and a bridge to self-protecting, smart B5G networks.

Anomaly-based cyberattack detection plays a crucial role in protecting smart home environments by detecting and preventing potential threats and unauthorized access. However, there is a discernible gap in evaluating existing datasets based on real-world smart home features, highlighting their use cases and limitations, and providing recommendations for developing more suitable datasets for dynamic smart homes. This paper includes a comparative analysis of some selected datasets, an assessment of how the datasets' limitations impact the performance of machine learning-based anomaly detection techniques, and a discussion of their implications for anomaly-based cybersecurity research and practice. Furthermore, this research serves as a foundation for future studies in smart home anomaly-based detection, emphasizing the importance of high-quality datasets and adaptive detection techniques in securing smart homes and protecting user privacy.

Online Social Networks (OSNs) have gained enormous popularity in recent years. They provide a dynamic platform for sharing content (text messages or multimedia) and for facilitating communication between friends and acquaintances. Microblogging services are a popular form of OSNs. They allow sending small messages in a one-to-many messaging model so that users can communicate with their favorite celebrity, brand, politician, or other regular users without the obligation of a pre-existing social relationship. A chain of privacy-related scandals linked to questionable data handling practices in microblogging services has arisen in the last past few years. Most current microblogging service providers offer centralized services and their business model is based on monitoring, analyzing, and selling users’ activity and patterns. In the end, the personal information shared by the users to benefit from the free-of-charge services is used for the underlying payment in such systems. In this paper, we present Garlanet, a privacy-aware censorship-resistant microblogging social network that does not rely on a centralized service provider as all data is hosted in computers voluntarily contributed by the users of the system. Garlanet provides microblogging functionalities while protecting privacy and preserving the confidentiality and integrity of users and data. It ensures that users’ identities and their social graphs are hidden from the system and adversaries and it provides availability and scalability of the services. We also evaluate the privacy level of Garlanet and we compare it with the privacy level of eight other microblogging systems.

Multi-Channel Man-in-the-Middle (MitM) attacks are special MitM attacks capable of manipulating encrypted wireless frames between two legitimate endpoints. Since its inception in 2014, attackers have been targeting Wi-Fi networks to perform different attacks, such as cipher downgrades, denial of service, key reinstallation attacks (KRACK) in 2017, and recently FragAttacks in 2021, which widely impacted millions of Wi-Fi devices, especially IoT devices. To the best of our knowledge, there are no studies in the literature that holistically review the different types of Multi-Channel MitM enabled attacks and analyze their potential impact. To this end, we evaluate the capabilities of Multi-Channel MitM and review every reported attack in the state of the art. We examine practical issues that hamper the total adoption of protection mechanisms, i.e., security patches and Protected Management Frames (PMF), and review available defense mechanisms in confronting the Multi-Channel MitM enabled attacks in the IoT context. Finally, we highlight the potential research problems and identify future research approaches in this field.

Online reservation systems have grown over the last recent years to facilitate the purchase of goods and services. Generally, reservation systems require that customers provide some personal data to make a reservation effective. With this data, service providers can check the consumer history and decide if the user is trustable enough to get the reserve. Although the reputation of a user is a good metric to implement the access control of the system, providing personal and sensitive data to the system presents high privacy risks, since the interests of a user are totally known and tracked by an external entity. In this paper we design an anonymous reservation protocol that uses reputations to profile the users and control their access to the offered services, but at the same time it preserves their privacy not only from the seller but the service provider.

Spectrum is an essential resource for the provision of mobile services. In order to control and delimit its use, governmental agencies set up regulatory policies. Unfortunately, such policies have led to a deficiency of spectrum as only few frequency bands are left unlicensed, and these are used for the majority of new emerging wireless applications. One promising way to alleviate the spectrum shortage problem is adopting a spectrum sharing paradigm in which frequency bands are used opportunistically. Cognitive radio is the key technology to enable this shift of paradigm.