Unmanned Aerial Vehicles (UAVs), or drones, are gaining popularity across several industries because of their adaptability, efficiency, and capacity to reach difficult-to-access regions. However, certain UAV applications carry critical or hazardous risks due to the environments in which they are used or the nature of the task. The main challenge associated with UAV communication is different types of cyberattacks. Hence, in this paper, a federated learning-based cyber-physical Intrusion Detection System (IDS) is proposed for UAV communication. A lightweight ANN model is embedded with each UAV as well as the central server. During learning, a chunk of data is used to train the local ANN models and, their weights are shared with the global model. The global model performs its weight updation and sends these weights back to each local model. The performance of the global model in terms of accuracy is compared against an isolated ANN model trained over the same data. An appreciable increase in the accuracy of the global model can be observed from this comparison. In addition, the performance of the local model and the global model is determined based on recall, precision, and F1-score.

Supply chain management (SCM) involves multiple distributed stakeholders and is increasingly exposed to security, transparency, and data integrity challenges due to extensive digitization. To address these concerns, this paper proposes a blockchain-enabled supply chain management framework that ensures secure, transparent, and tamper-resistant transaction processing across the supply chain network. The proposed system records transaction data generated by key stakeholders in the supply chain, such as manufacturers, wholesalers, distributors, retailers, and customers, and stores them in blocks. Each transaction is verified for validity; if valid, it proceeds to further processing. These blocks are mined to confirm the consensus mechanism. Once validated, they are added to the blockchain. A smart contract automatically releases payment upon successful delivery of the items. The framework has been implemented in MATLAB to simulate a multi-node supply chain environment. Performance evaluation is conducted using metrics such as mining duration, nonce behavior, and consensus latency across different Proof-of-Work difficulty levels to assess the efficiency of the blockchain-based SCM. Overall, the proposed framework highlights the feasibility of integrating blockchain technology into SCM to enhance trust, traceability, and operational transparency while providing insights into consensus-aware performance trade-offs.

Cardiovascular disease (CVD) is one of the most common and major global health challenges, which requires improved methods for early and precise detection and intervention. So, to recognize these heart problems and avoid sudden cardiac arrest, it is essential to detect abnormal heart conditions early. Machine learning (ML) based medical treatments are being implemented that are very helpful in quickly and effectively diagnosing CVD problems. One method that can offer practical answers to these kinds of problems is a meta-heuristic approach. Owing to its effectiveness, meta-heuristic approaches are presently used with medical data to diagnose conditions more practically and successfully than the traditional ML methods. In this study, we used three different meta-heuristic algorithms which are Genetic Algorithm (GA), Cuckoo Search Algorithm (CSA) and Particle Swarm Optimization (PSO) for diagnosis of the CVD diseases using two different datasets - CVD and Framingham. Finally, various ML classifiers were applied on the best selected features for both the datasets, obtained from the meta-heuristic algorithms for finding efficiency and comparing the results. The results demonstrate that Framingham dataset gives best accuracy of 98.47% by using CSA algorithm for feature selection and Random Forest as classifier whereas for the CVD dataset gives best accuracy of 94.12% by using PSO algorithm and Random Forest as classifier. Then, the best performing model is passed through some fuzzy logic rules to improve the model accuracy and gives better prediction for CVD prediction.

Industry 5.0 represents a paradigm shift toward human-centric, intelligent, and sustainable manufacturing systems. At the core of this transformation lies the Digital Twin (DT), a virtual replica of physical assets that enables real-time monitoring, simulation, and decision-making. This article presents a comprehensive meta-analysis of how DT technologies contribute to the realization of Industry 5.0 objectives across domains such as Smart Additive Manufacturing (SAM), Predictive Maintenance (PM), Cyber-Physical Cognitive Systems (CPCS), Intelligent Supply Chain (ISC), and Adaptive Scheduling (AS). By analyzing 125 peer-reviewed studies, we quantify the feature-wise attainment levels of Industry 5.0 and identify critical gaps in current implementations. The findings reveal that, while SAM exhibits the highest Industry 5.0 readiness, other features, such as cognitive systems, remain underdeveloped. The article concludes by outlining key research challenges and presenting a strategic roadmap to advance the real-world integration of DTs within Industry 5.0 frameworks.

Due to the growing quantity of medical images, the patient data recognition is the important one for treatment clinical institutions. So, digital image watermarking has widespread importance at the time of sharing of medical images among treatment clinical institutions. But in watermarking the extraction of the patient data and preserving image quality are the major important factors, after sharing of medical images. In this paper, a metrics achieved in theproposed scheme have better perfomance or novel spatial domain-based reversible digital image watermarking scheme proposed on medicalcompressionimagesandtoensuredecompressionauthenticity,integrity,watermarkingandmedicalconfidentiality.imagesInthisscheme, the input ground truth image (or cover image) is divided into 4 X 4 block sizes. The watermark image Keywords: Medical image Watermarking, Reversible watermarking, Spatial domainScalarquantizationCompressionDecompressionQRcode is embedded with the original ground truth image by performing L-level Quantization. The watermark image is quick response (QR) code of patient information. The compression applied to the remainder part of the cover ground truth image using the mean arithmetic method to make the process most robust. The reversible watermarking scheme applied over watermarked images by performing inverse L-level Quantization and decompression. The proposed scheme is compared with Reza et al. existing scheme: "An Image Watermarking Algorithm for Medical Computerized Tomography Images". The proposed scheme has better experimental results compared to the existing scheme in terms of Peak Signal-to-noise ratio (PSNR), Mean Square Error (MSE), and Structural Similarity Index (SSIM). The PSNR, MSE, and SSIM metrics are used to evaluate the robustness of the proposed scheme for extracting cover images and QR code of patient information (in terms of watermark images) from watermarked images. In addition, the Similarity (SIM) and the correlation coefficient (CRC) are used to evaluate how well the suggested approach for patient data and cover ground truth medical image extraction holds up robustness. The experimental numerical valuesof metrics achieved in the proposed scheme have better performance for compression and decompression watermarking medical images.

The paper focuses on handwritten digit detection using the MNIST 0-9 dataset to develop an automated grading system. Four distinct machine learning models are employed: basic and latest Neural Network Models, such as Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN), as well as traditional models like K-Nearest Neighbors (KNN) and Gradient Boosting (XGBoost). The objective is to accurately identify handwritten digits, which is crucial for grading purposes. By evaluating the models' performance metrics, the aim is to determine the most suitable model for efficient digit detection. The ultimate goal is to streamline the grading process, ensuring timely result declaration while maintaining proper record maintenance.

With the exponential growth of Internet of Things (IoT) devices, edge computing has emerged as a vital paradigm for localized data processing, low-latency communication, and efficient resource utilization. However, routing in IoT-based edge networks remains challenging due to dynamic topologies, constrained energy resources, and growing security threats. To tackle these challenges, this article proposes QLB-IoT, a novel framework that integrates Q-learning-based intelligent routing with blockchain-assisted trust management to optimize energy consumption while ensuring secure and adaptive data transmission. The proposed method enables IoT devices to autonomously learn energy-efficient routing paths based on real-time network parameters such as residual energy and link distance. The Q-learning mechanism minimizes route flapping and promotes adaptive decision-making. Simultaneously, blockchain and digital signatures establish a decentralized trust layer that authenticates devices and prevents attacks like blackhole and Sybil. Compared with open shortest path first (OSPF), Q-Routing, deep reinforcement learning-based control framework for traffic engineering (DRL-TE), and intelligent edge network routing (ENIR), exhibits an extended network lifetime to approximate to 1300 rounds (2.4x OSPF and 8% beyond ENIR), increases throughput by 18%-30%, and reduces average latency by half, while minimizing packet loss by approximate to 40% and increasing the likelihood of quality of service (QoS) compliance by up to 10%. These improvements stem from a real-time exploration-exploitation mechanism that dynamically reroutes traffic away from energy-depleted nodes, and from the immutable ledger that prevents malicious route manipulation without centralized oversight. By combining adaptive, energy-aware routing with verifiable trust, QLB-IoT delivers a scalable and resilient solution for mission-critical IoT-edge deployments.

The proliferation of Cyber-Physical Systems (CPS) in precision agriculture has led to the generation of massive spatio-temporal datasets across distributed sensors and edge devices. While Federated Learning (FL) offers a privacy-preserving framework for collaborative model training, its application in Agricultural CPS (Agri-CPS) remains limited due to high communication costs, heterogeneous data distributions, and computational constraints of edge hardware. To address these challenges, we propose SecureLightCrypt—a novel FL framework that integrates lightweight compressed encryption with spatio-temporal learning and High Performance Computing (HPC)-driven simulation. The framework employs sparse homomorphic masking, ternary quantization, and delta compression to reduce communication overhead by up to $10 \times$ without compromising accuracy. Additionally, HPC-assisted simulations generate auxiliary features and weak labels from climate and pest dynamics, augmenting local training performance. Extensive experiments on real-world crop yield and disease datasets demonstrate that SecureLightCrypt outperforms state-of-the-art FL baselines in accuracy, efficiency, and robustness while ensuring data privacy and scalability. This work bridges the gap between secure FL and simulation-informed agricultural intelligence, enabling next-generation smart farming applications.

Rice leaf disease classification employing machine learning approaches is an important and ongoing research area due to its wider production and consumption across the globe. Although it is evident that Convolutional Neural Networks (CNNs) have brought about a significant paradigm change in the domain of image recognition, especially when detecting agricultural diseases. However, its ‘black box’ nature prevents humans from understanding and interpreting its work for decision making. In this context, it is preferable to make an effort to render it explainable. Hence, in this paper, the working of the CNN model is explained through three important techniques such as layer-wise relevance propagation (LRP), SHapley Additive exPlanations (SHAP), and Local Interpretable Model-agnostic Explanations (LIME). A modified version of LRP is used to explain the workings of each CNN layer by propagating the relevance score backwards over the network. The model prediction is analyzed and explained using SHAP. The region of importance that impacts the model to predict the specific disease is explained using LIME. This work provides a clear vision of the functioning and interpretation of the CNN model for different classifications of rice leaf diseases. Experimental results found 96.5

Cardiovascular disorders (heart diseases) are the most prevalent cause of death on a global scale. So early detection and classification increase the likelihood of survival. In the context of machine learning techniques, there is always a need for an accurate and explainable predictive model for detecting various diseases, such as cardiac disorders. The work carried out in this paper stacks bidirectional long short-term memory with deep learning to propose two models. The first model is used to detect cardiac disease with a binary label classification, while the second one classifies cardiac disease, which is a multi-label classification problem. Bidirectional LSTM is used as an approximate algorithm for feature extraction. Deep learning is used for classification purposes. The proposed models are trained and validated over the PTB-XL dataset. The performance of these models is evaluated and compared against state-of-the-art methods. The comparison shows the proposed model outperforms other methods in terms of accuracy, precision, f1-score, and recall. SHAP is used to make these models explainable, which in turn helps to annotate different diseases on the ECG report.

This study takes a closer look at how quantum computing can be used to run regression models, a common type of machine learning algorithm. Our goal is to understand whether today's quantum systems—ranging from open-source tools like Qiskit to commercial platforms such as AWS Braket, Microsoft Azure Quantum, and IBM Quantum—can effectively support these models in practice. To explore this, several quantum libraries and kernel methods are used to implement and test different versions of regression algorithms. Then compared the performance of these quantum models with traditional, classical ones to see how they stack up in terms of speed, accuracy and practical usability. Simulated results offer a balanced view of where quantum computing stands today: showing both its promise and the areas where it still has room to grow. Overall, this work aims to shed light on the real-world potential of quantum technology in machine learning applications.

This paper proposes a multi-step stock forecasting framework built on Deep Operator Networks (DeepONet). The proposed architecture pairs a dense feedforward “trunk” that models the structure of future price trajectories with an LSTM-based “branch” that learns temporal patterns from historical quotations. Using the previous 30 days as input, the system generates 10-day-ahead forecasts. Because it utilizes neural operators, the model identifies intricate patterns and remains effective even when market conditions change. In simulations, the approach maintains accuracy in the presence of noise and outliers. Also consistently surpasses standard time-series baselines, achieving lower mean squared error (MSE) and root mean squared error (RMSE). These results demonstrate DeepONet's strong potential for practical stock market prediction and its effectiveness in modeling the dynamics of financial time series

Cyber-Physical Systems (CPS) in healthcare, such as wearable and implantable medical devices, require realtime, privacy-preserving, and personalized intelligence to support adaptive monitoring and control. However, traditional cloudbased learning frameworks are limited by latency, bandwidth constraints, and privacy risks. This paper presents a novel framework, HPC-Augmented Federated Meta-Learning (HFML), which integrates High-Performance Computing (HPC) resources with Federated Meta-Learning to address these limitations. In the proposed system, HPC infrastructure is employed to meta-train a global initialization model across diverse patient data cohorts. This meta-model is then distributed to CPS devices, enabling rapid on-device adaptation through few-shot local updates tailored to individual patients. The framework enhances scalability, personalization, and robustness while preserving data privacy. Simulation-based evaluation demonstrates that HFML achieves faster convergence and improved accuracy on non-IID, patientspecific data compared to conventional FL approaches. The proposed architecture offers a promising direction for deploying intelligent, adaptive control in CPS-based healthcare applications.

The security of Internet of Medical Things (IoMT) devices is crucial for ensuring the integrity and reliability of patients’ medical data. These devices, operating over TCP and ICMP protocols, are highly susceptible to cyberattacks. While machine learning models can detect these attacks with acceptable accuracy, their operational mechanisms remain unclear, leaving the decision-making process of the models undefined. Moreover, the accuracy and training time of machine learning models is more questionable when the datasets has large number of sparse features and class imbalances. This study introduces an interpretable feature selection technique designed to enhance intrusion detection in IoMT by reducing redundant features and improving model efficiency. Random Forest-based explainable AI model provides transparency in attack classification and better decision-making. The simulated results employing the CICIoMT2024 dataset demonstrate that the proposed method significantly improves detection performance, with the Random Forest model achieving 99% accuracy, outperforming XGBoost (98%), Decision Tree (97%), and Support Vector (98%), while ensuring explainability through SHAP-based feature analysis. Thus, the simulation outcomes reveal the key contributing factors for various cyberattacks on IoMT, facilitating enhanced security measures and real-time monitoring. The proposed approach boosts detection accuracy and interpretability, making it highly suitable for real-world IoMT security applications.

Semantic aerial imagery segmentation is of great importance in land use classification, urban planning and disaster estimation. Its capability is compromised by limitations such as a lack of enough labeled data and data authenticity and security issues. In this paper, we present a novel framework that combines transfer learning with blockchain technology to break such limitations. Transfer learning enables the usage of pretrained deep neural networks that reduce the necessity for large sets of labeled data without sacrificing segmentation performance. For integrity, traceability, and secure sharing of imagery information and segmentation outputs, a blockchain infrastructure is implemented in the framework. The distributed database securely retains model output and metadata, enabling multiple stakeholders to collaborate without jeopardizing data privacy. Experimental validation verifies that the proposed method performs segmentation best with enhanced training efficiency and secure data management. Through the combination of the latest machine learning and distributed technology, the framework offers a flexible and reliable solution for geospatial analysis, particularly in reliable data-critical applications like disaster management and smart city planning.

Integrating Federated Learning (FL) with Cyber-Physical Systems (CPS) is a novel approach to sustainable agriculture. CPS uses real-time data from distributed sensors, drones, and edge devices to monitor and manage important agricultural elements. FL allows farmers to train collaborative learning without exchanging raw data, securing confidentiality and privacy. In several key application areas, such as precision irrigation to minimize water waste, yield prediction for effective resource allocation, pest and disease detection to minimize the need for pesticides, smart crop rotation to preserve soil fertility, and adjusting to climates for proactive planning. CPS and FL work together to produce sustainable outcomes. We present FedAgri, a federated, sensor-aware learning algorithm designed for resource-constrained and diverse agricultural environments that can achieve these objectives. FedAgri uses client selection, adaptive learning, and model compression techniques to improve scalability and reduce communication overhead. By addressing the issues of data management and environmental sustainability, the proposed framework promotes efficient, adaptable, and intelligent agricultural practices.

Dynamic routing problem is an important research challenge for the robust working of the Internet of Things network as many sensors are equipped with limited energy resources. The work carried out in this paper uses a novel energy model to formulate the routing problem for IoT networks. An AI-based approach using a modified A* algorithm proposes a dynamic routing framework for IoT. The simulated results are compared with the k-Means algorithm and genetic algorithm in terms of energy consumption and the number of dead nodes. This comparison reveals the proposed routing framework to outperform than k-Means algorithm and GA by extending the lifetime of the IoT network by a factor of thrice and twice, respectively. Further, the proposed routing framework is also compared over different network configurations.