Abstract Predictive maintenance approaches to improve naval processes and streamline shipbuilding developments conducted during the inner working at Hindustan Shipyard in Visakhapatnam. Moreover, the study overcomes many problems like intricate complex operational environments and data excellence. Basically, the key parameters such as loads, speeds, torques, and vibrations were observed by using the intelligent cutting-edge sensors. To predict the performance of the equipment, various Machine Learning and Deep Learning models were enabled such as Bi Directional Long Short Terms Memory Network with Spatial Pyramid Attention Networks (BiLSTM-SPAN) and Random Forest. Consequently, BiLSTM-SPAN model outperformed well with a performance by attaining of R 2 = 0.910 and significantly lower error rates. The outcomes are highlights that the developed BiLSTM-SPAN model successfully enhances the operational efficiency, accordingly, decreasing the downtime, and enhance maintenance practices, thereby making at the shipyard more competitive operations through the application of innovative intelligent predictive strategies.

Abstract Nowadays, the leading causes of early mortality, especially in countries with low or middle incomes are cardiovascular diseases (CVDs), which include heart failure, stroke, hypertension, and coronary artery disease. Premature death rates may be reduced if certain illnesses are identified early. Numerous methods, including data mining and machine learning (ML) have been proposed by researchers for early identification as well as tracking of cardiac patients in the context of CVD prediction. Yet there remains significant concern about the effectiveness of these strategies in scenarios where there is a high error rate and uncertain precision. Thereby, choosing a prediction method that may produce high accuracy and less errors is essential. This article presents the Squeeze-improved convolutional neural network (ICNN) model for detecting CVD via electrocardiogram (ECG) images. During preprocessing, the ECG image noise is reduced using an improved median filter (IMF). The model extracts shape, deep, and local Gabor XOR pattern (LGXP) features. For classification features, SqueezeNet and improved CNN (IMP-CNN) are combined within a hybrid model architecture (Squeeze-ICNN), with a Comb SigHyper-Hsine activation function in the classifier. The model provided experimental results proving accuracy (95.6%), precision (97.9%), sensitivity (94.9%), specificity (92.2%), and lower false negative rate (FNR) (5.1%) than conventional models (e.g., convolutional neural network [CNN], deep neural network [DNN], LSTM, DenseNet). The study clearly proves the effectiveness and performance of the model to detect CVD from ECG images.

Abstract Diabetic retinopathy (DR) is an eye issue of prolonged diabetes, which continues to be a major cause of vision loss among middle-aged adults. Early detection of DR through advanced imaging and artificial intelligence (AI) techniques can lead to a significant reduction in risk and severity of visual loss. This proposed ARMDiaRD combines state-of-the-art techniques from EfficientNet, Swin Transformers, and multi-scale feature fusion (MSFF) to enhance multiclass classification of DR severity levels across diverse generalised datasets. In the proposed architecture, a global context (GC) block is integrated into EfficientNet to capture long-range dependencies and contextual relationships. This is followed by Swin Transformer layers equipped with an MSFF block that hierarchically aggregates features from multiple levels, enabling the model to learn richer and more discriminative representations. The fundus images of four publicly available datasets for DR: asia pacific tele ophthalmology society (APTOS), Indian diabetic retinopathy image dataset (IDRID), MESSIDOR-V1, and EYEPACS, are combined through a comprehensive aggregation process to facilitate the training and testing of the generalised model. This design demonstrates consistent improvements across multiple evaluation metrics, underlining its potential to reduce misclassification in medical diagnosis. To evaluate the proposed model, the performance metrics, such as accuracy, precision, recall, specificity, and F1-score, are calculated along with quadratic weighted kappa (QWK), Spearman rank correlation coefficient, and mean absolute error (MAE). Simulation results revealed that the proposed model achieved an accuracy of 87.59%, precision of 87.6%, recall of 87.9%, QWK of 91.47% and Spearman coefficients with 92.53% respectively. Importantly, the MAE, a critical metric for evaluating false predictions in medical diagnosis, is 0.1736 in the proposed model. The outcome clearly demonstrates the supremacy of the proposed model over the other models in handling multiple larger datasets. This research shows the ubiquitous nature of the proposed model to predict the severity grades using larger complex datasets. This proposed ARMDiaRD confirms that combining local, global, and hierarchical characteristics is highly effective in preventing overfitting issues with the existing architecture, such as CNN, when tested using new data for reliable clinical image analysis.

Abstract Accurate identification of weed and crop species is essential for implementing precision agriculture practices that minimize chemical usage and improve yield. This study presents an advanced machine learning framework that employs optimized classification techniques to distinguish between multiple plant species based on image features. Utilizing a dataset comprising 11,500 images from diverse agricultural classes—including crops such as maize and sugar beet, and weeds like thistle and wild oat—the framework extracts color, texture, and shape features for high-resolution pattern recognition. Several supervised machine learning algorithms, including eXtreme gradient boosting, light gradient boosting machine, and ensemble stacking methods, are explored and fine-tuned using feature selection techniques such as recursive feature elimination (RFE) and principal component analysis (PCA). The experimental results demonstrate superior performance, with the best models achieving classification accuracy of up to 96.1% and reduced inference times suitable for real-time agricultural applications. These findings highlight the potential of optimized machine learning pipelines in advancing sustainable weed management and automated field monitoring.

Abstract This study introduces a novel algorithm for determining elastic constants in materials using a multiresolution approach. This innovative method accurately estimates the natural frequency of the tested material, a crucial parameter for evaluating material constants such as Young’s modulus and stiffness. This algorithm has substantial significance in the field of materials characterization. The experimental setup involves a specialized mechanical assembly designed for testing metal plates. Vibrational signals are acquired during experimentation. These raw signals undergo a multi-resolution process, leading to the extraction of the material specimen’s fundamental frequency. Using this estimated fundamental frequency, essential elastic constants such as Young’s modulus and stiffness are calculated with high precision. To validate the experimental findings, a thorough comparison is made using the finite element method (FEM), specifically utilizing ANSYS software. The experimental procedure is conducted on two different materials: stainless steel SA 240 Gr 304 and copper. Additionally, various testing conditions, including changes in ball weight and ball release height, are employed to comprehensively assess the algorithm’s performance across diverse scenarios. The experimental outcomes highlight the algorithm’s accuracy. Notably, the average percentage errors for estimating the fundamental frequency are impressively low, standing at 1.30% for stainless steel (SS) and 1.19% for copper. Similarly, the algorithm demonstrates excellence in calculating the modulus of elasticity, resulting in average percentage errors of 1.97% for SS and 2.33% for copper.

Abstract Current energy management faces its most significant power optimization struggle because rising demand meets with environmental pressures and requires effective resource utilization. The research develops an advanced implementation of bitterling fish optimization (BFO) combined with the genetic algorithm (GA) to perform power consumption optimization in intricate systems. The proposed framework connects BFO swarm intelligence to GA evolutionary strategies through machine learning adaptation, which simultaneously minimizes power losses and improves system efficiency throughout multiple applications. The hybrid approach produces quicker convergence through improved optimization performance and delivers flexibility toward dynamic energy requirements. The model requires consideration of three vital elements: load distribution management and renewable integration, and power loss reduction, along with interactive visual controls to monitor power distribution and optimization progress, and system output. Additionally, cloud and edge computing technologies enable large-scale data processing and ensure low-latency optimization for real-time applications. The framework also supports adaptive learning mechanisms, allowing it to continuously refine optimization strategies based on real-time energy data and consumption patterns. The proposed power optimization framework has wide-ranging applications in smart grids, industrial automation, and renewable energy systems, offering a scalable, high-accuracy solution for modern energy management. The integration of predictive analytics and real-time monitoring empowers decision-makers with proactive energy-saving strategies, ultimately contributing to a sustainable, efficient, and intelligent power infrastructure. This research provides a novel, data-driven approach to power optimization, addressing critical challenges in energy management while enhancing reliability and operational efficiency in future energy infrastructures.

Abstract The growing demand for privacy-preserving machine learning has driven the advancement of federated learning (FL), facilitating decentralized model training without revealing raw data. Nevertheless, conventional FL methods frequently encounter challenges in achieving an optimal balance between privacy, efficiency, and accuracy. To address these challenges, this study introduces the hybrid adaptive learning framework (HALF)—an innovative, scalable, and privacy-focused FL approach. HALF implements a multi-tiered hierarchical structure that includes local, regional, and global aggregation, combined with adaptive learning rates and integrated differential privacy (ε ≤ 2.3), ensuring robust privacy protection without sacrificing performance. The framework employs a structured five-phase process: Initialization, Device Selection, Training, Aggregation, and Evaluation, and incorporates lightweight algorithms with efficient communication protocols to reduce latency, energy consumption, and overhead. Experimental validation using benchmark datasets demonstrates that HALF consistently surpasses traditional FL by 15%–20% in accuracy while reducing communication overhead by 40%, maintaining ≥89.7% accuracy within 142 min of training. Its resilience under non-identical and independently distributed (non-IID) and heterogeneous data conditions, as well as its deployment with up to 10,000 clients, underscores its scalability. Relevant to critical sectors such as healthcare, Internet of Things (IoT), finance, and smart cities, HALF addresses existing FL limitations and establishes a foundation for secure, efficient, and adaptable AI solutions in sensitive data environments.

Abstract To address the challenge of quantitatively identifying mixed gases, we developed a gas concentration identification algorithm based on the sparrow search algorithm (SSA) and optimized support vector regression (SVR). The Tent chaotic mapping operator is employed to initialize the population, enhancing population diversity, and improving the algorithm’s global search capability. By optimizing SVR parameters with SSA, we propose an enhanced TSSA-SVR model. Evaluated on mixed gas datasets, TSSA-SVR achieves a prediction accuracy of 94.47%, outperforming comparative algorithms such as Genetic Algorithm (GA)-SVR and PSO-SVR, while demonstrating improved convergence compared to the baseline SSA-SVR. The experimental results demonstrate significant performance enhancements, offering an effective solution for precise gas concentration identification in complex environments.

Abstract The applications of drones for smart farming are well accepted nowadays. It also results in huge fiscal losses to the agricultural economy. In conventional agriculture, resources are wasted due to the constant and uniform use of pesticides, fertilizers, and pharmaceuticals. Nevertheless, within the existing literature, no comprehensive approach to these difficulties has been discovered. The current research uses drones to irrigate and distribute insecticides, fertilizers, and medicine to necessary crops in sufficient quantities. By understanding current needs, it generates and stores enough synthetic data. Through optimal resource usage and synthetic data production and analysis, it boosts agro-profit.

Abstract The study focuses on intelligent driving, emphasizing the importance of recognizing nearby vehicles and estimating their positions using visual input from a single image. It employs transfer learning techniques, integrating deep convolutional networks’ features into a modified CenterNet model for six-degrees-of-freedom (6DoF) vehicle position estimation. To address the vanishing gradient problem, the model incorporates simultaneous double convolutional blocks with skip connections. Utilizing the ApolloCar3D dataset, which surpasses KITTI in comprehensiveness, the study evaluates pretrained models’ performance using mean average precision (mAP). The recommended model, Center-DenseNet201, achieves a mAP of 11.82% for relative translation thresholds (A3DP-Rel) and 39.92% for absolute translation thresholds (A3DP-Abs). These findings highlight the effectiveness of pretrained models in the modified architecture, enhancing vehicle posture prediction accuracy from single images. The research contributes to autonomous vehicle development, fostering safer and more efficient navigation systems in real-world traffic scenarios.