Data transmission via laser beams using Free Space Optics can be achieved with high-speed wireless communication between line-of-sight transmitters and receivers, and has rapid deployment, extreme network security, and does not require any license-based bandwidth, resulting in affordable communication. However, atmospheric conditions, especially rain, significantly affect the performance of FSO. This research explores the impact of rainfall on FSO systems in Malta’s Mediterranean environment, which is marked by moderate rain. Several widely used rain attenuation models are analyzed and validated against Malta’s rainfall data. The study also investigates wavelength dependence (850 nm, 1050 nm, and 1550 nm), identifying 1550 nm as the most reliable for minimizing rain-induced attenuation. Referring to the records of past rainfalls in Malta, the research concludes that the Suriza model with a wavelength of 1550 nm is the most appropriate choice for ensuring FSO communications. The findings are vital in improving the reliability and efficiency of such FSO systems in similar climates in the Mediterranean region and can be implemented to investigate other climates where rain plays a major role in signal degradation.
Millions of lives are lost every year due to cardiovascular disease (CVD), making it a global health crisis. This presents an unprecedented opportunity for the development of accurate, early diagnostic solutions. A huge number of healthcare data is frequently going underutilized, resulting in missed chances to identify trends that might help in prompt diagnosis. This study discusses machine learning methods that can be utilized in CVD prediction and save millions of lives at risk. This research pinpoints the optimal revolutionizing cardiovascular diagnostics approaches via the assessment of various machine learning models including Voting Classifier, Light GBM, Auto ML TPOT, Neural Networks, and others, to revolutionize CVD prediction by utilizing underused healthcare data. The results reveal that the Voting Classifier (73.47
The increasing development of quantum computing is a serious threat to most classical cryptographic algorithms such as RSA and ECC (Elliptic Curve Cryptography), which are the current basis for securing mobile wallet transactions. With the threat of quantum computing powered attacks that will safely disable existing public-key schemes, we provide a provably secure and forward-thinking solution in a contribution to the area of mobile wallets with a post-quantum cryptography (PQC) timeline, specifically, a dynamic PQC key update algorithm for mobile wallet security. This method introduces the ML-KEM (Module Lattice–Key Encapsulation Mechanism), a lattice-based algorithm that NIST has started standardization on, along with the use of AES symmetric encryption, to form a multi-layer security framework. Essentially two infrastructure (keys) are used, Data Encryption Keys (DEK’s) that encrypt transaction’s data, and Key Encryption Keys (KEK’s), that Per Springer style, both city and country names must be present in the affiliations. Accordingly, we have inserted the city and country names “Oxford, UK” in the affiliation. Please check and confirm if the inserted city and country names are correct. If not, please provide us with the correct city and country names.wrap DEK’s with an encapsulation using ML-KEM, to create a hybrid encryption with post-quantum resistance. A dynamic rolling-key system enhances the architecture by renewing the DEK’s and KEK’s at fixed intervals, providing forward secrecy, while also withholding any potential damage from compromising keys. This solution remains backward compatible to ensure legacy data can still be decrypted after the key is updated. Through this double-layer encryption and dynamic key management approach, the system achieves post-quantum resistance, low computational overhead, and seamless adaptability, offering a practical pathway to securing mobile wallet transactions against both current and future cryptographic threats.
Artificial intelligence (AI) is increasingly deployed in urban health ecosystems to promote active lifestyles and improve wellbeing. However, the environmental cost of resource-intensive models and ethical concerns around personalization highlight the need for Green and Responsible AI frameworks. This paper presents Lively-Vida (Li-Vida), a mobile system that delivers personalized exercise recommendations through lightweight Human Activity Recognition (HAR) and on-device collaborative filtering. By minimizing reliance on cloud infrastructure, Li-Vida achieves energy efficiency, reduced environmental footprint, and privacy preservation. Behavioral nudging strategies (transparent reminders and achievement badges) are embedded and evaluated using the Theory of Reasoned Action (TRA). A 30-day empirical study with 17 university students demonstrates statistically significant improvements in attitudes (ATT), subjective norms (SN), and behavioral intentions (BI) toward physical activity (all p < .001; very large effect sizes). Qualitative analysis of open responses shows that transparent nudges reinforced motivation, progress visibility, and social accountability. Findings demonstrate that Green and Responsible AI can produce measurable health behavior change in a smart-city context, offering a scalable path to sustainable urban wellbeing.
Underwater wireless optical communication (UWOC) systems rely on laser-based transmission to achieve high data rates, but their performance is strongly affected by physical impairments. In this paper, we investigate the effect of beat noise and turbulence on the detected intensity of overlapping laser signals. The analysis considers multiple lasers operating at the same frequency with random phase and polarization angles, which give rise to beat noise, while turbulence is modeled using the Weibull distribution to represent strong underwater conditions caused by salinity variations and air bubbles. Results show that beat noise is the dominant factor influencing the statistical distribution of the received signal, particularly as the number of overlapping lasers increases. However, turbulence further broadens the probability density function (PDF), leading to higher variance and reduced signal stability.
Indoor localization and navigation continue to present significant research challenges due to the unreliability of Global Positioning System (GPS) in indoor environments. In this paper, we present a hybrid IoT-enabled indoor positioning framework that amalgamates Wi-Fi fingerprinting, machine learning-driven coordinate regression, and adaptive path planning. The Received Signal Strength Indicator (RSSI) fingerprints access points, a Large Language Model (LLM)-based agent for data cleaning, and a Random Forest classifier and regressor are used to determine the room ID and the exact coordinates of the user on the floor, respectively. A Kalman filter is employed to mitigate noise and environmental instabilities, while Dijkstra’s algorithm is implemented for obstacle-aware indoor navigation. The framework is deployed in a university campus, achieving a Mean Average Error (MAE) of 0.37 m in coordinate estimation, and over 99
This study creates and uses a strong machine-learning framework to detect chronic kidney disease (CKD) early by using a publicly available dataset of 400 patient records and 25 clinical and laboratory markers. We addressed issues with data quality by utilising k-nearest neighbours imputation for missing values, SMOTE oversampling to rectify a 62.5/37.5 class imbalance, one-hot encoding for categorical variables, and z-score normalisation for continuous predictors. We used grid search with five-fold cross-validation to tune seven supervised algorithms: Support Vector Machine, Decision Tree, Random Forest, K-Nearest Neighbors, XGBoost, LightGBM, and a deep Feedforward Neural Network. The Neural Network had the best regression fidelity (R2 = 0.9729, MSE = 0.0031), finding complex nonlinear correlations between biomarkers. LightGBM, on the other hand, exhibited flawless discrimination (ROC-AUC = 1.00, precision = 1.00, recall = 0.967) with minimal additional computing power required. The accuracy of Random Forest and K-Nearest Neighbors was good but not as good as the others (0.975 and 0.9625, respectively). To build clinical trust, we used Shapley Additive Explanations (SHAP), which always found that specific gravity, albumin, and hemoglobin were the three most important factors in predicting CKD risk. Models were serialized in ONNX format and made available through a secure, Docker-containerized Flask API. This made it possible for hospital IT systems to do real-time inference and show risk for each patient. The single-center dataset and absence of longitudinal data constrain generalizability; nonetheless, prospective multi-center validation, federated learning, and multi-modal data integration are advised to enhance these findings. Our method combines advanced machine learning, strict preprocessing, and explainable AI to make AI-enhanced nephrology possible. This can help find CKD early, make better use of resources, and improve patient outcomes.
Blockchain technology has rapidly evolved from its cryptocurrency origins into a promising solution for securing sensitive domains such as healthcare. However, as healthcare blockchain networks integrate diverse medical devices, they are increasingly exposed to cyber-physical attacks that target device integrity and compromise patient data. Detecting such malicious devices is challenging, as false positives can disrupt essential clinical processes. This research proposes a novel framework that combines supervised Machine Learning (ML) with blockchain smart contracts to enhance the security of healthcare blockchain networks. A synthetic Ethereum-based dataset is generated to simulate healthcare transactions, which is then used to train five ML classifiers: Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest, Decision Tree, and Gradient Boosting Machine (GBM). Among these, the Decision Tree algorithm achieved the best performance, offering higher accuracy with minimal false positives. To further address false alarms, the framework introduces a multifaceted re-authentication mechanism embedded in smart contracts. When a device is flagged as suspicious, the reauthentication process validates its legitimacy before restricting or restoring its functionality. This dual-layered defence not only detects compromised devices effectively but also ensures uninterrupted operation for legitimate ones. The integration of ML anomaly detection with blockchain re-authentication represents a proactive strategy for safeguarding healthcare networks against evolving cyber-physical threats.
The rise of smart cities has made artificial intelligence (AI) an important part of urban life. Yet, deploying AI in these systems creates a tension between enhancing efficiency and protecting human autonomy and independence. Traditional behavioral economics introduce the concept of nudges, which are known to be clear, straight forward, direct interventions that guide choices while preserving human freedom. In contrast, AI-driven nudges are highly personalized, adaptive, and are often unnoticed by users, raising complex ethical concerns about their influence on daily decision-making. To address this, we propose the BEHAVE framework (Behavioral Ethics for Human-Aware AI in Virtual Environments), which integrates sustainable AI, responsible machine learning, and human-centered design for smart cities. Through systematic analysis of smart city implementations, we identify three behavioral paradoxes: the cognitive load paradox, transparency trade-off, and sustainability rebound effect. The framework introduces six guiding principles: Behavioral Transparency, Ethical Timing, Human Agency Preservation, Adaptive Consent, Vulnerability Protection, and Evaluation Continuity to navigate these challenges. A case study in urban traffic management illustrates how ethical AI nudging can effectively promote sustainability while ensuring users retain meaningful control over their choices.
The proliferation of mobile services demands authentication protocols that remain secure against quantum adversaries while preserving user privacy and minimizing computational overhead. This paper presents a post-quantum cryptography (PQC) based mutual authentication and session-key establishment protocol for roaming mobile users interacting with foreign and home agents. The design employs lightweight hash-and-XOR constructions with fresh nonces to achieve mutual authentication, forward secrecy, and resistance to replay and impersonation attacks in resource-constrained environments. Security is validated through both informal analysis and formal verification using the AVISPA tool with the OFMC backend, confirming secrecy of the negotiated session key and robustness against active intruders. Implementation and performance evaluation demonstrate low computational and communication costs compared to existing schemes, making the protocol suitable for next-generation mobile networks requiring quantum-resilient security and user anonymity.
The goal of this paper is to provide a new telecommunications transport network management model that will be adequate for the modern and future Beyond 5G telecommunications networks. The paper focuses on such a matching problem and defines a method to calculate probabilistic-temporal characteristics, including for tunnels through terrestrial and non-terrestrial network nodes. Thus, the scheme proposed in this article offers new traffic management capabilities based on transport network types, namely: optical, satellite, and DOCSIS cable networks. AI and ML models for traffic engineering in such networks are also considered.
This research details the development and validation of a Smart Construction Helmet augmented with AI, IoT sensors, and edge computing. The helmet is equipped with a multi-sensor array on a Raspberry Pi 4 platform, running a YOLOv5s model for real-time Personal Protective Equipment (PPE) detection and an optimized LSTM for predictive hazard analytics. The system was rigorously evaluated in a laboratory setting designed to mimic construction site conditions. Experimental results show the system attains 88.7
This article presents the development of a meteorological data collection and transmission system utilizing wireless sensor networks (WSNs) based on LoRa technology. The study examines approaches to message routing in sensor networks, including centralized and decentralized algorithms, as well as methods for enhancing network fault tolerance and optimizing energy consumption. A self-organizing network algorithm, based on energy metrics such as device battery level, is proposed to optimize routing paths. The work also investigates a mechanism for configuring radio module parameters to reduce the energy costs associated with data transmission. Simulation results demonstrate a significant reduction in energy consumption, confirming the proposed system's effectiveness for real-world applications.
This paper analyzes modern approaches and methods for estimating resource consumption in fog computing systems. Based on this analysis, a new method for estimating and forecasting computing resources using machine learning algorithms is proposed and theoretically substantiated. A prototype load monitoring and forecasting system was developed in Python using Docker containers. The developed system was tested and its performance was compared with existing approaches in a number of key metrics. Experiments confirmed the effectiveness of the proposed method: increased accuracy in node load forecasting, reduced request processing latency, and prevented system overload, improving overall service quality.
This paper presents a scalable Ultra-wideband (UWB) radar based system integrated with a deep convolutional neural network (CNN) for multi-human detection and tracking in indoor environments. Using over 6,400 radar samples, the data is processed through a multi-step signal processing pipeline, further classified by a 2D Convolutional Neural Network (CNN) which accurately classifies four occupancy states(empty to three persons), with 99.2
Free Space Optical Communication (FSO) has emerged as a dependable alternative to traditional wireless communication, offering the additional benefits of higher data rates and secure transmission. However, FSO systems are susceptible to atmospheric conditions and regional climate, especially in areas with snowy weather, such as Norway and other parts of the world, where signal loss can occur due to poor visibility. This study quantifies how snowfall-induced visibility variations affect the signal-to-noise ratio (SNR) and link reach in snow-prone regions using an artificial neural network (ANN) model. The ANN model achieved an excellent fit between predicted and actual SNR values. The simulations indicate a steep decrease in SNR as visibility decreases and a heavy snowfall (visibility 200 m or less) rapidly reduces SNR to below 0 dB, making long-range linking infeasible. Accordingly, snowfall significantly degrades FSO performance in Northern European countries and other snow-prone regions, whereas Malta’s mild climate offers favourable conditions for FSO deployment. This work presents an ANN-based method for estimating SNR across visibility classes to support feasibility assessment and performance planning for FSO links in snowy environments.
Ransomware and malware pose serious risks to cloud system resilience and cybersecurity, as the scale and complexity of data render traditional detection systems increasingly inadequate. Advancements in AI and ML technologies show strong potential in addressing these challenges, offering improved accuracy, adaptability, and automation. This paper presents a comprehensive review of the state-of-the-art AI- and ML-driven approaches for identifying ransomware and malware threats in cloud ecosystems. The review introduces a taxonomy of detection methods, encompassing traditional, machine learning, deep learning, hybrid, explainable, and privacy-preserving approaches. It further evaluates widely used performance metrics and benchmark datasets, highlighting their strengths, limitations, and suitability for reproducible research. A comparative summary of representative studies demonstrates both the promise of AI-driven detection and the fragmentation of current approaches to this field. Key gaps remain in resilience against adversarial inputs, explainability, privacy-preserving learning, IoT and edge-cloud scalability, and standardised benchmarking. To address these gaps, the paper proposes a research agenda that emphasises the development of robust and explainable models, federated and privacy-preserving paradigms, IoT-specific ransomware datasets, and integration with natively cloud-integrated infrastructures, alongside stronger ethical and legal safeguards. By synthesising current knowledge and outlining recommended directions for future study, this paper contributes a structured roadmap for advancing resilient and trustworthy AI-driven ransomware and malware detection in cloud environments.
For a long time honeypots have been an effective cyber defense capability to divert attackers to virtual environments where you can learn on emerging threats. For cloud computing, the traditional honeypots are restricted from scaling and agility for advanced zero-day exploits that are complex in nature. Artificial Intelligence (AI) honed deception mechanisms have made their way into honeypot systems we really are killing two birds communicating with one stone, by integrating AI within ethical hacking and cloud security right now a days. In this paper, we will offer a survey of AI-enabled honeypots applied in code honeypots as well as distinctions made between conventional honeypot approaches and latest AI-powered work, such as machine learning based on detection, adaptive response mechanisms and also generative deception techniques along with ethical issues like deception legitimacy, consent for use of personal data and risk of entrapment. We devise a theory of honeypot detection and AI robustness for the implementation of ethical AI honeypots in cloud systems to show some theoretically intractable challenges of honeypot detectability and AI robustness can be tackle. Finally, we highlight the open problems and future directions to steer the further evolution of honeypot systems with the ability to predict advance zero-day attacks in an cloud environment.
The rapid expansion of Artificial Intelligence (AI) presents a paradox: while it drives innovation and economic growth, it simultaneously intensifies environmental and sustainability challenges, particularly through its dependence on energy-hungry data centres. Ireland has emerged as a major European hub for data centres, generating both economic optimism and societal concern. This paper investigates how Irish news media has framed discourse around the growth of data centres in the context of AI-driven demand and national sustainability commitments. Using topic modelling on a corpus of Irish news articles published between July 2024 and August 2025, we identify and analyse the most prominent themes shaping public and policy debates. Our findings highlight recurring tensions between economic development, energy security, and environmental sustainability, while also revealing how the accelerating deployment of AI is increasingly recognised as a compounding challenge. By surfacing these narratives, this study contributes to understanding the social and political dynamics underpinning Ireland’s contested role as Europe’s data center hub. We conclude by discussing the implications for aligning AI growth with green energy transitions and national policy coherence.
The integration of Artificial Intelligence (AI) into the medical field is revolutionizing healthcare by enabling efficient diagnostics and decision-making. A key challenge in medical AI is to ensure accurate predictions while maintaining data privacy and confidentiality. Several works have tried to address this challenge using Federated Learning (FL), but most of them are limited to simulated environments with no real-world implementation. We propose a real-time Websocket communication-based FL method for classifying medical images, which enables several hospitals to work together to train a deep learning (DL) model without disclosing private patient information. Each hospital, acting as a worker node, trains a local DL model on medical image datasets. Locally trained models generate weight parameters that are encrypted using homomorphic encryption (HE). These encrypted weights are then safely transmitted to a centralized master node, where they are aggregated to create an updated global DL model. The final model is then deployed back to the worker nodes, ensuring improved accuracy and generalization. Communications between the worker nodes and the master node are facilitated using WebSocket-based connections. Using FL and encryption techniques, this real-time implementation enhances model performance while preserving patient data confidentiality, demonstrating the practical feasibility of AI-driven secure healthcare solutions.
