Advances in microelectronics and the evolution of IoHT devices allow the collection and, consequently, generation of a greater volume of health data, intensifying the need for robust data privacy solutions. Traditionalk-anonymity-based anonymization techniques often suffer from high information loss, especially as the anonymity parameter k increases. To address these challenges, this article proposes Generalization Firstk-Member Clustering (GFKMC), a novelk-anonymity method that applies early generalization to quasi-identifiers, reducing computational overhead and minimizing information loss. Unlike traditional methods (e.g., Mondrian, Top-Down Greedy (TDG), and Clustering-Based (CB)), GFKMC maintains nearly constant information loss (≈ 25%) across varyingkvalues and better preserves machine learning model performance, especially in lowkscenarios. Empirical evaluations demonstrate that GFKMC outperforms baseline methods by significantly minimizing the trade-off between data utility and privacy. Moreover, GFKMC preserves the performance of machine learning models more effectively.

The increasing volume of sensitive data generated by various domains demands robust approaches to privacy protection. Anonymization based on k-anonymity stands out for mitigating the risks of re-identification of personal data. However, the impact on the performance of machine learning models is commonly neglected. This work proposes a novel comparative method to evaluate the effects of anonymization on the performance of machine learning models, considering privacy, information loss, and performance metrics. The empirical results show how generalization based on k-anonymity impacts federated learning solutions and provides insights for developing and improving methods that reconcile data privacy and efficiency.

The growing prevalence of interconnected devices in the Internet of Things (IoT) has intensified concerns about data security and user authentication. This study investigates a multimodal biometric authentication approach integrating Photoplethysmography (PPG) and Electroencephalography (EEG) signals to enhance accuracy and robustness. We adapted a previously validated PPG and Electrocardiogram (ECG) model, replacing the ECG component with an EEG-based method optimized through hyperparameter tuning. Our findings demonstrate that combining brain and heart signals improves authentication performance, with the multimodal approach surpassing unimodal methods. The optimized model achieved a precision of $97.05 \%$, a recall of $97.19 \%$, a F1-Score of $96.96 \%$, and an accuracy of $97.05 \%$, highlighting the potential of EEG-PPG fusion in secure biometric authentication.

Technological advances enable the integration of Internet of Things (IoT) devices to perform continuous and proactive patient monitoring. These devices collect a large volume of sensitive data that requires privacy. Anonymization provides privacy by removing or modifying information that identifies an individual. However, traditional anonymization techniques, such as k-anonymity, depend on a fixed and pre-defined k value, susceptible to attribute disclosure attacks. This article presents Dynamic Anonymization by Separatrices (DAS), an approach for defining the ideal value k and for dynamic grouping of data to be anonymized using separatrices measurements. Results show that the proposed approach efficiently mitigates attribute disclosure attacks.

In an increasingly connected world, ensuring security in e-health is a challenge. Traditional security models based on perimeter trust are insufficient to guarantee the protection of these systems. Since these models work by directly assigning trust to the user, the entire network becomes vulnerable if the user’s credentials or device are compromised. Thus, this work proposes and evaluates a model based on Zero Trust to considerably increase security in e-health environments. The proposed model uses privilege reduction and user confidence analysis to perform access control. The evaluation follows simulation in different scenarios, assessing their assertiveness in delegating access. The results show the effective detection of anomalies in accesses by the model.

The 5G technology is an evolution of the mobile networks due to the adoption of a Service Based Architecture (SBA). It allows greater flexibility and the possibility of using data analysis processes in network management. The 3GPP defines the Network Data Analytics Function (NWDAF) as the network function responsible for data analysis in 5G, however, this function remains considerably unexplored in the existing literature. In order to fill this gap, this paper investigates the NWDAF using a dataset from a simulated 5G network and employing machine learning models as network protocol classifiers. The results achieved an accuracy up to approximately 75% while the average lies around 65%, providing not only reproducibility but also paving the way for future investigations in data analysis as specified by the 3GPP.

With the significant increase in mobile users connected to the wireless network, coupled with the escalating energy consumption and the risk of network saturation, the search for resource management has become paramount. Managing several access points throughout a whole region is hugely relevant in this context. Moreover, a wireless network must keep its Service Level Agreement, regardless of the number of connected users. With that in mind, in this work, we propose four prediction models that allow one to predict the number of connected users on a wireless network. Once the number of users has been predicted, the network resources can be properly allocated, minimizing the number of active access points. We investigate the use of Particle Swarm Optimization and Genetic Algorithms to hyper-parameterize a Multilayer Perceptron neural network and a Decision Tree. We evaluate our proposal using a campus-based wireless network dataset with more than 20,000 connected users. As a result, our model can considerably improve network performance by intelligently allocating the number of access points, thereby addressing concerns related to energy consumption and network saturation. The results have shown an average accuracy of 95.18%, managing to save network resources effectively.

Although the 5G New Radio standard empowers the mobile communication networks with diverse technologies such as Massive MIMO, mmWave deployments, and much more, some network functionalities still do not explore the potential of assembling Artificial Intelligence to their methodologies. The handover procedure is planned very similarly to in older 3GPP networks, based on simple power measurement comparisons and rudimentary parameter tuning, such as Time-To-Trigger and Hysteresis. This work develops and evaluates with simulations and real network data a new Deep Learning approach to support the handover triggering decision toward a data-driven procedure for next-generation networks. Our solution relies on predicting future samples of standard Reference Signals using Long Short-Term Memory Networks (LSTM) in the first stage. After, the predicted power samples are sent to a binary classification algorithm to identify if the time series will lead or not to a handover triggering. The results show a mean absolute error of around 0.6 dB predicting power signal samples and over 97% of accuracy, indicating the future handover trigger moment. Finally, we discuss possible use cases to implement our model, including Open RAN and MEC architectures.

The performance of Attribute-Based Access Control is negatively affected by communications over the network between the policy decision point and policy information point for each attribute request. Attribute caching is a standard solution to mitigate this problem. However, due to the dynamic nature of attributes, the cost to keep them refreshed increases for each new attribute replica. This paper presents a method that predicts each request and anticipate the positioning of the attributes closer to the requester exploring the tradeoff between the cost of creating a new replica versus updating a replica. The proposed method follows a two-way approach, where it deals with the current attribute requests and their estimates based on user mobility and the best positions for the attributes. Through trace-driven simulations, considering traces from a large university campus, the method shows a reduction of up to 80% in the number of hops to get the needed attributes at negligible refreshment cost.

Over the years, wireless networks have been suffering a significant increase in the number of connected users, and dealing with this increase is extremely important in terms of economy and quality of service. In this work, a prediction model is proposed to improve this relationship, focusing on predicting the number of connected users to the wireless network. Our model consists of a particle swarm optimizer applied to the parameters of the Multilayer Perceptron neural network. The model was evaluated with real mobility data obtained from wireless networks with a total of more than twenty thousand users. The predictions made by the model allow allocating the network bandwidth efficiently, generating savings in the available resources. In fact, the simulated results indicate an average coefficient of determination of 94.08% and average savings of 67.31 % of the total available bandwidth.

Network technologies for large areas based on subGHz have emerged as a way to provide long-range communication with low cost and complexity. Among the various existing solutions, LoRa is arguably the most adopted and promising in this context. Its main application has been allowing the possibility of ubiquitous connectivity to IoT from a simple network and management structure. Some factors must be taken into account when proposing a LoRa network. The type of application directly affects the LoRa network communication, such as the center frequency, the spreading factor, the bandwidth, and the coding rates chosen by each node. We will observe in this work the characteristics of LoRa physical-layer and its automatic configuration based on the quality of the perceived signal-to-noise ratio (SNR). In this way, we propose an adaptable protocol for LoRa networks with low overhead and complexity. The results obtained by a real scenario show that only 23% of the observation time make changes in the configuration and has an average gain of 4.68% for SNR.

With current advances in the deployment of testbeds for Future Internet (FI), a new challenge arises: identity management in a globally distributed environment. In this context, it is necessary to understand local and federated models of identity management to integrate testbeds. This paper presents the design and implementation of a module for credential translation that enables a user of an academic authentication and authorization (A&A) federation, such as CAFe (the Brazilian Federated Academic Community), to access the FI testbed federation. The proposed model supports the integration of testbed federations and academic federations. The proposal generates X.509 certificates and other standard credentials used in the testbed federation, following the SFA standard, based on user attributes obtained from the A&A federation (CAFe). The developed module also allows an attribute-based access control, denying or allowing a user access according to his/her attributes obtained from CAFe. Other contributions are based on facilities for the user to delegate his SFA credential to an experimenter control interface. The study was conducted using a real experimentation laboratory (GIDLab), in which mirrors of the CAFe federation and of the MySlice platform were set up to allow the comparison of security features of our scheme to other proposals.