Punch-through accidents pose a significant risk during the positioning of jack-up rigs. To mitigate this hazard, accurate prediction of the peak penetration resistance of spudcan foundations is essential for developing safe operational plans. Advances in artificial intelligence have spurred the widespread application of machine learning (ML) to geotechnical engineering. To evaluate the prediction effect of different algorithm frameworks on the peak resistance of spudcans, this study evaluates the feasibility of ML and multi-view learning (MVL) methods using existing centrifuge test data. Six ML models—Random Forest, Support Vector Machine (with Gauss, second-degree, and third-degree polynomial kernels), Multiple Linear Regression, and Neural Networks—alongside a Ridge Regression-based MVL method are employed. The performance of these models is rigorously assessed through training and testing across various working conditions. The results indicate that well-trained ML and MVL models achieve accurate predictions for both sand-over-clay and three-layer clay strata. For the sand-over-clay stratum, the mean relative error (MRE) across the 58-case dataset is approximately 15%. The Neural Network and MVL method demonstrate the highest accuracy. This study provides a viable and effective empirical solution for predicting spudcan peak resistance and offers practical guidance for algorithm selection in different stratigraphic conditions, ultimately supporting enhanced safety planning for jack-up rig operations.

Accurate fault diagnosis of power transformers is critical for maintaining grid reliability, yet conventional dissolved gas analysis (DGA) methods face challenges in feature representation and high-dimensional data processing. This paper presents an intelligent diagnostic framework that synergistically integrates systematic feature engineering, tensor decomposition-based feature selection, and a sparrow search algorithm (SSA)-optimized multi-kernel support vector machine (MKSVM) for transformer fault classification. The proposed approach first expands the original five-dimensional gas concentration measurements to a twelve-dimensional feature space by incorporating domain-driven IEC 60599 ratio indicators and statistical aggregation descriptors, effectively capturing nonlinear interactions among gas components. Subsequently, a novel Tucker decomposition framework is developed to construct a three-way tensor encoding sample–feature–class relationships, where feature importance is quantified through both discriminative power and structural significance in low-rank representations, successfully reducing dimensionality from twelve to seven critical features while retaining 95% of discriminative information. The multi-kernel SVM architecture combines radial basis function, polynomial, and sigmoid kernels with optimized weights and hyperparameters configured through SSA’s hierarchical producer–scrounger search mechanism. Experimental validation on DGA samples across seven fault categories demonstrates that the proposed method achieves 98.33% classification accuracy, significantly outperforming existing methods, including kernel PCA-based approaches, deep learning models, and ensemble techniques. The framework establishes a reliable and accurate solution for transformer condition monitoring in power systems.

Purpose: Sentiment analysis is an important aspect of understanding consumers' views on products, especially in the growing skincare industry. This study aims to compare the accuracy and effectiveness of various kernels in the Support Vector Machine (SVM) algorithm, including linear, polynomial (poly), and radial basis function (RBF) kernels, in predicting three types of sentiment: positive, neutral, and negative based on reviews of local Indonesian skincare products.Design/methodology/approach: The dataset used includes consumer reviews classified by rating, which are then processed using Term Frequency-Inverse Document Frequency (TF-IDF) technique for feature extraction.Findings/result: The evaluation results show that the RBF kernel achieves the highest accuracy of 74.78%, followed by the linear kernel with 74.51% accuracy, and the polynomial kernel with 74.10% accuracy. Although the difference between the three kernels is not significant, the RBF kernel excels in positive sentiment classification, while all three kernels struggle in predicting neutral and negative classes.Originality/value/state of the art: These findings make an important contribution to the development of effective sentiment analysis methods, especially in the context of datasets with high class imbalance. To handle class imbalance, techniques such as oversampling smaller classes or using cost-sensitive learning techniques to give more weight to negative and neutral classes can be used.

To ensure the reliable operation of power converters and prevent catastrophic failures, this paper proposes a novel online fault diagnosis strategy for a four-level converter. The core of this strategy is an optimized multi-kernel extreme learning machine model. Specifically, the model extracts multi-scale features from three-phase output currents by combining Gaussian and polynomial kernels and employs particle swarm optimization to determine the optimal kernel fusion scheme. Experimental validation was performed on an online diagnosis platform for a four-level converter. The results show that the proposed method achieves a high diagnostic accuracy of 99.35% for open-circuit faults. Compared to conventional methods, this strategy significantly enhances diagnostic speed and accuracy through its optimized multi-kernel mechanism.

Modern crop recommendation systems must accurately grasp the complex and nonlinear relationships between soil nutrients to support effective agricultural decisions. In this study, we introduce a framework that combines supervised and unsupervised learning through kernel feature fusion, integrating Radial Basis Function (RBF) Kernel Principal Component Analysis (KPCA) and Kernel Linear Discriminant Analysis (KLDA) into a single seven-dimensional embedding. First, six principal components are extracted using RBF-KPCA to capture global nonlinear variance in the raw data. Similarly, in the raw space, an Nystroem-approximated RBF transformation followed by LDA produces a single discriminant axis (KLDA) for better supervised class separation. These features are fused by concatenation and then input into Support Vector Machine (SVM) classifiers (using polynomial and RBF kernels) and a Random Forest (RF) classifier. In the experiments, a publicly available dataset comparing maize and barley based on six soil features was used. The fused representation significantly outperformed raw data and single-embedding methods, with Polynomial SVM increasing by 18.5%, RBF SVM improving by 10.1%, and RF rising by 4.7% over the raw data. These results show that combining unsupervised variance maximization with supervised discriminant projection creates a richer, more discriminative feature space—especially beneficial for SVMs in crop recommendation tasks. Our kernel fusion approach offers a powerful and flexible strategy for precision agriculture, enabling robust decision support without extensive field trials or repeated laboratory tests.

Rice is a major staple crop that is highly susceptible to various leaf diseases, necessitating an accurate early detection method to prevent yield losses. This study proposes a hybrid approach combining Convolutional Neural Network (CNN) and Support Vector Machine (SVM) for rice leaf disease classification based on digital images. The CNN is employed as a deep feature extractor, while the SVM serves as the main classifier. The dataset consists of rice leaf images categorized into four disease types: Bacterial blight, Blast, Brown spot, and Tungro. The data were divided into training and validation sets, and the CNN model was trained for 10 epochs, achieving a validation accuracy of 98.14% at the 10th epoch. The extracted CNN features were then evaluated using different SVM kernels, namely Linear, Polynomial, RBF, and Sigmoid. The experimental results show that the Sigmoid kernel achieved the best performance with an accuracy of 49%, followed by Polynomial, RBF, and Linear kernels.

Understanding nuclear shape, behavior, and stability, as well as improving nuclear models, depends on the precise determination of ground-state nuclear charge radii. Existing experimental techniques are limited to extremely narrow regions of the nuclear chart; theoretical models, including relativistic Hartree-Bogoliubov (RHB) and Hartree-Fock-Bogoliubov (HFB), predict broad trends of nuclear properties but miss fine isotopic features such as odd-even staggering effects and shell-closure kinks. High computational time and cost are other obstacles to theoretical approaches. Although machine-learning algorithms have made significant progress in predicting charge radii, they are still hindered by a lack of balanced data and characteristics, primarily centered around and . In the present study, we present the first application of CatBoost regression to compute nuclear charge radii. We integrated two experimental datasets with RHB-calculated point-coupling interaction (PC-X) theoretical features and extended our study range to , . We found the best hyperparameters using Optuna’s Tree-structured Parzen Estimator (TPE) sampler with 10-fold cross-validation (CV), achieving a CV root-mean-square error (RMSE) of 0.0106 fm and hold-out RMSE of 0.0102 fm, with only three features, i.e., neutron number (N), proton number (Z), and RHB theoretical binding energy (BE), outperforming nine other ML models: random forest (RF), quantile RF (QRF), Cubist, Gaussian process regression with polynomial kernel (GPPK), multivariate adaptive regression splines (MARS), SVR, ANN, convolutional neural network (CNN), and Brussels-Skyrme-on-a-grid 3 (BSkG3). SHapley Additive exPlanations (SHAP) analysis confirms the highest global influence of BE in the model's predictions, followed by proton number and neutron number. The proposed model can accurately reproduce the kink and odd-even staggering effects in krypton and strontium chains. These results establish CatBoost as a robust and notably promising model for charge-radius prediction and beyond, with the potential to impact r-process modeling and future theoretical development.

<p>Egg production and consumption in Indonesia continue to rise, highlighting the need for accurate egg quality assessment. This study evaluated egg quality using a Support Vector Machine (SVM) model that integrates image and non-image features through feature-level fusion. A total of 750 eggs were analyzed based on external characteristics (shell color, cleanliness, texture, weight, and images) and internal characteristics (odor, albumen, yolk, black spots, images). Image data were reprocessed through grayscale conversion, resizing, and texture extraction using the Gray Level Co-occurrence Matrix (GLCM). Both linear and polynomial SVM kernel with varying degrees were tested, and the polynomial kernel (degree 6) achieved the best, with 86% accuracy, 91% precision, and 87% recall. These results demonstrate that integrating image and non-image features significantly enhances egg quality classification compared to using either data type alone. These findings provide valuable insights for developing automated egg grading system in the poultry industry.</p>

Background:Bipolar disorder (BD) and attention deficit hyperactivity disorder (ADHD) are two distinct psychiatric disorders characterized by significant overlap in symptoms, making differential diagnosis challenging. Due to the lack of a definitive test for diagnosing and differentiating these disorders, the present study aimed to accurately diagnose and differentiate between patients with BD and ADHD using the support vector machines (SVM) with radial basis function, polynomial, and mixture kernels, as well as ensemble neural networks, to analyze functional magnetic resonance imaging (fMRI) data.Materials and Methods:In this study, 49 individuals with BD and 40 individuals with ADHD were analyzed. A protocol based on fMRI imaging and a switching task was proposed for diagnosing ADHD and BD. The graph theory method calculated the graph criteria using the CONN toolbox in 15 areas of the attention circuit. The effective features were then selected using the genetic algorithm (GA), and finally, the performance of the models was evaluated using four criteria: accuracy (ACC), sensitivity (SE), specificity (SP), and area under the curve (AUC).Results:57 effective and important features were selected as input features by GAs with 99.78% ACC. The performance score of the models showed that the SVM with mixture kernels model performed best among the other algorithms (ACC = 92.1%, SE = 92.6%, SP = 97.3%, and AUC = 0.931).Conclusion:According to the evaluation criteria values, the best model for diagnosing ADHD from BD has been suggested. This approach can be useful in diagnosis, psychological, and psychiatric interventions.

This study conducted 537 force-controlled compression tests on shoe soles to measure strain energy. Various features were studied, including midsole and outsole structures, sole geometry, sole hardness, and loading conditions including applied compression force, angle, and loading speed. Strain energies during compression tests were recorded, followed by correlation analysis, analysis of variance (ANOVA), and Tukey tests. Subsequently, different machine learning (ML) algorithms—i.e., simple linear regression (SLR), support vector regression (SVR), random forest (RF), and multi-layer perceptron (MLP)—were applied for curve fitting. The correlation matrix revealed that strain energy exhibits a strong dependence on the applied force (correlation = 0.89), which stands out significantly compared to other features. This is followed by geometry and shoe hardness, both of which show a correlation of approximately 0.25. To validate these findings, ANOVA tests were conducted, and all p values were significant (p < 0.05). Additionally, it was found that auxetic midsole structures achieve similar strain energy values as simple midsoles (mean = 2.978 J), while maintaining a lighter overall weight. In contrast, weight-reducing holes were not as effective (mean = 2.080 J). It was also shown that initial contact position (i.e., changed with the compression angle) can affect the strain energy; that is, initial contact with the heel generates less strain energy (mean = 2.643 J) compared to the first touch with the toes (mean = 2.839 J). After tuning ML hyperparameters through grid search, RF outperformed other models (MSE = 0.0089), followed by SVR with polynomial kernel (MSE = 0.0105), SVR with RBF kernel (MSE = 0.0146), and MLP (MSE = 0.0251). The developed models offer a practical tool for shoe manufacturers to optimize sole designs and prevent injury during high-stress activities.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-20864-3.

Modern dental education increasingly calls for smarter tools that combine precision with meaningful feedback. In response, this study presents the Intelligent Dental Handpiece (IDH), a next-generation training tool designed to support dental students and professionals by providing real-time insights into their techniques. The IDH integrates motion sensors and a lightweight machine learning system to monitor and classify hand movements during practice sessions. The system classifies three motion states: Alert (10°-15° deviation), Lever Range (0°-10°), and Stop Range (>15°), based on IMU-derived features. A dataset collected from 61 practitioners was used to train and evaluate three machine learning models: Logistic Regression, Random Forest, Support Vector Machine (Linear RBF, Polynomial kernels), and a Neural Network. Performance across models ranged from 98.52% to 100% accuracy, with Random Forest and Logistic Regression achieving perfect classification and AUC scores of 1.00. Motion features such as Deviation, Take Time, and Device type were most influential in predicting skill levels. The IDH offers a practical and scalable solution for improving dexterity, safety, and confidence in dental training environments.

This study devised a fog recognition model and simulated traffic flow in low visibility. It initially built a cloud image recognition model based on convolutional neural network and support vector machine. Subsequently, a mixed traffic flow model was developed for low-visibility conditions. The results showed that the Gaussian kernel function achieved the highest fog image recognition accuracy, reaching 92.58%, while the polynomial kernel function had the lowest accuracy of 84.19%. When five experiments were conducted, the fog image recognition model in this study exhibited the highest accuracy (0.94), recall (0.875), and F1 score (F1) (0.9). In a vertical driving formation, vehicles ahead travelled faster, indicating that the convergence speed and stability of the full speed difference within the formation were improved. The enhanced intelligent driver model demonstrated minimal speed fluctuations, with all vehicles in an 8-car fleet reaching a stable driving speed within 40 s. This implies excellent stability of the improved intelligent driver model. In conclusion, the model developed in this research shows promising practical applications in fog recognition and traffic flow management under low visibility, and has positive significance for improving highway safety performance.

In response to the challenge of dynamic adaptability in operational safety assessment for UAVs operating in complex operational environments, this study proposes a novel operational safety assessment method based on an Improved Support Vector Machine. An operational safety assessment index system encompassing four dimensions—operator, UAV platform, flight environment, flight mission—is constructed to provide a comprehensive foundation for evaluation. The method introduces a dynamic weighted information entropy mechanism based on a sliding window, overcoming the static features and delayed response of traditional SVM methods. Additionally, it integrates Gaussian and polynomial kernel functions to significantly enhance the generalization capability and classification accuracy of the SVM model in complex operational environments. Experimental results show that the proposed model demonstrates superior performance on test samples, effectively improving the accuracy of operational safety assessment for the Reconnaissance–Strike UAV in complex operational environments, and offering a novel methodology for UAV safety assessment.

Foreign language skills are one of the gateways to opening up opportunities for better education and employment. The use of technology can help with this, especially handwriting recognition technology. However, the use of limited datasets is often a problem. This study uses a Convolutional Neural Network (CNN) model with a Residual Network (ResNet) architecture and a Support Vector Machine (SVM). ResNet, as a feature extraction method for the data, is capable of capturing data patterns without losing much of the original data information. Meanwhile, the SVM algorithm, as a data classifier, is capable of working well with limited data. This research uses hyperparameters of linear kernel, polynomial kernel, Radial Basis Function (RBF) kernel, and Sigmoid kernel. Additionally, the hyperparameters C and Gamma values were also used. The research results indicate that the best model accuracy was obtained from the model trained with a linear kernel and a C value of 0.1, with an accuracy of 81.72% and an accuracy on the test data of 87.50%.

Sleep disorders such as insomnia and sleep apnea are health problems that can have a serious impact on a person's quality of life. Early detection of these disorders is important to prevent the risk of more severe complications. This study aims to build a sleep disorder classification model using the Support Vector Machine (SVM) algorithm by evaluating the influence of four types of kernels, namely Linear, Polynomial, Radial Basis Function (RBF), and Sigmoid. The dataset used comes from the Sleep Health and Lifestyle Dataset which contains information about individual characteristics related to sleep and lifestyle. The research process follows the CRISP-DM stages from data understanding, data preparation, modeling, to model evaluation using accuracy, precision, recall, and F1-score metrics. The evaluation results show that the Polynomial kernel produces the best performance with 91.6% accuracy, followed by Linear, RBF, and Sigmoid. This finding shows that the selection of the right kernel in SVM has a significant effect on classification quality. This research contributes to the utilization of machine learning to detect sleep disorders and opens up opportunities for the development of more accurate and efficient diagnostic systems.

Water quality and quantity affect crop productivity, with surface water quality having a significant impact. The amount of surface water being used for drinking is gradually rising. Thus, assessing surface water quality and related hydro-chemical characteristics is essential for surface water resource management in Mahanadi River Basin, Odisha. The current study examined surface water quality and appropriateness for drinking and agriculture, utilizing several techniques such as Weighted Arithmetic (WA) Water Quality Index (WQI), Multivariate models namely Pearson Correlation, Cluster Analysis (CA) and Principal Component Analysis (PCA), six multiple machine learning (ML) techniques like, gaussian process regression (GPR), linear regression (Stepwise), fit binary tree (FBT), support vector regression, SVM (linear and polynomial kernels), and artificial neural network (ANN) to predict the WQI, for sustainable use of the surface water resources. Thirteen physicochemical parameters were used to analyse eleven surface water samples, which indicating that the primary cation and anion concentrations were as follows: Mg2+ > Ca2+ > K+ > Na+, and HCO3- > Cl- > SO42- > NO3-, respectively. The best input combination for WQI model prediction was identified using subset regression analysis. These eight input combinations had high R2, ranging from 0.975 to 1, and high Adjusted R2 amounts to 0.974-1. The WAWQI range is divided into five categories: excellent (18.18%), good (18.18%), poor (27.27%), very poor (27.27%), and unsuitable (9.09%). The study discovered that increased turbidity concentration, carbonate weathering, and the growth of agricultural and urban-industrial sectors regulate the geographical variance in surface water quality. The correlation results depict that the significant positive correlation has been found between TDS and TH (0.87), Mg2+ with turbidity (0.84) and coliform (0.78), Ca2+ and coliform (0.72), Cl- and HCO3- (0.83), and K+ and Na+ (0.7). Owing to the correlation study, these ions are enriched in the surface water by major anthropogenic activity. While, in the present study, CA and PCA has been used to determine the surface water's governing factors. Differentiation of three clusters based on the sources, hydrogeochemical environment, and reactions between chemical variables by utilizing CA and the results of PCA shows that the first three primary components (PCs) account for 84.76% of the overall variation. Hence, CA and PCA shows the several processes that are the main sources of the ions, such as carbonate, silicate weathering, and evaporate dissolution. Pursuant to the stepwise fitting model, bicarbonate was a non-significant variable for the WQI, whereas turbidity, pH, and coliform were the most significant factors. With a high correlation of 1 and low errors, the results demonstrated that the GPR, stepwise linear regression, and ANN models outperformed the others during the training and testing phases.In contrast, during the training and testing stages, the SVM and FBT models showed the lowest performance. Therefore, the GPR, stepwise regression, and ANN models exhibited low mistakes and a strong correlation during the training and testing phases. In conclusion, the combination of physicochemical characteristics, WQI, CA, PCA, and ML tools to assess the surface water suitability for drinking and irrigation and their regulating variables are beneficial and provides a clear picture of water quality. Future research should improve the data accuracy to increase model precision and extend its applicability to various geographical and environmental settings.

Background: Food allergies represent a growing global health concern, yet the current diagnostic methods often fail to distinguish between true allergies and food sensitivities, leading to misdiagnoses and inadequate treatment. Epigenetic alterations, such as DNA methylation (DNAm), may offer novel biomarkers for precise diagnosis. Methods: This study employed a computational machine learning framework integrated with DNAm data to identify potential biomarkers and enhance diagnostic accuracy. Differential methylation analysis was performed using the limma package to identify informative CpG features, which were then analyzed with advanced algorithms, including SVM (polynomial and RBF kernels), k-NN, Random Forest, and artificial neural networks (ANN). Deep learning via a stacked autoencoder (SAE) further enriched the analysis by uncovering epigenetic patterns and reducing feature dimensionality. To ensure robustness, the identified biomarkers were independently validated using the external dataset GSE114135. Results: The hybrid machine learning models revealed LDHC and SLC35G2 methylation as promising biomarkers for food allergy prediction. Notably, the methylation pattern of the LDHC gene showed significant potential in distinguishing individuals with food allergies from those with food sensitivity. Additionally, the integration of machine learning and deep learning provided a robust platform for analyzing complex epigenetic data. Importantly, validation on GSE114135 confirmed the reproducibility and reliability of these findings across independent cohorts. Conclusions: This study demonstrates the potential of combining machine learning with DNAm data to advance precision medicine in food allergy diagnosis. The results highlight LDHC and SLC35G2 as robust epigenetic biomarkers, validated across two independent datasets (GSE114134 and GSE114135). These findings underscore the importance of developing clinical tests that incorporate these biomarkers to reduce misdiagnosis and lay the groundwork for exploring epigenetic regulation in allergic diseases.

Diabetes mellitus is a chronic disease that occurs due to excessively high blood glucose levels resulting in the absence of insulin. In the period of data at the Siti Khadijah Islamic Hospital in Palembang, which is influenced by the number of patients undergoing health checks such as diabetes mellitus, it affects the classification of data that will complicate the hospital. So by utilizing data mining, classification to determine patients who have undergone examinations including diabetes sufferers or not. With these problems, the author conducted a comparative analysis of two algorithms, namely the naïve Bayes algorithm and the support vector machine algorithm for the classification of diabetes by using the WEKA tool with the Cross Validation and Confusion Matrix options tools with the highest accuracy results, namely the support vector machine algorithm with a polynomial kernel, the results of which are 96.2704% and an error rate of 3.7296%, it can be concluded that the most accurate algorithm in the classification of diabetes is the support vector machine algorithm with a polynomial kernel.

This study developed a novel Water Quality Index (WQI) using Kernel Principal Component Analysis (PCA) to assess groundwater quality (GWQ) in the coastal aquifers of Al-Qatif, Saudi Arabia. A total of 39 groundwater samples were collected from shallow and deep wells and analyzed for key physicochemical parameters. Six kernel types were tested, and the polynomial kernel was found to be most effective in preserving variance and reducing dimensionality. The Kernel PCA-based WQI classified wells into ‘Very Bad,’ ‘Bad,’ and ‘Medium’ categories, with scores such as W3 (WQI = 25.51, “Very Bad”), W31 (WQI = 46.7, “Bad”), and W38 (WQI = 56.75, “Medium”). Salinity and EC presented poor Sub-Index (SI) scores, reflecting the impact of seawater intrusion and over-extraction, while pH consistently showed high SI values (100), indicating natural buffering. By integrating non-linear dimensionality reduction, the proposed framework enhances traditional WQIs and facilitates more targeted and transparent groundwater decision-making. This includes identifying priority wells for remediation and supporting sustainable abstraction policies. The findings offer insight into sustainable water management in arid and semi-arid regions that are confronting groundwater degradation.

Cycle packing is a fundamental problem in optimization, graph theory, and algorithms. Motivated by recent advancements in finding vertex-disjoint paths between a specified set of vertices that either minimize the total length of the paths [Björklund and Husfeldt, ICALP 2014; Mari et al., SODA 2024] or request the paths to be shortest [Lochet, SODA 2021], we consider the following cycle packing problems: Min-Sum Cycle Packing and Shortest Cycle Packing . In Min-Sum Cycle Packing , we try to find, in a weighted undirected graph, \( k \) vertex-disjoint cycles of minimum total weight. Our first main result is an algorithm that, for any fixed \( k \) , solves the problem in polynomial time. We complement this result by establishing the W[1]-hardness of Min-Sum Cycle Packing parameterized by \( k \) . The same results hold for the version of the problem where the task is to find \( k \) edge-disjoint cycles. Our second main result concerns Shortest Cycle Packing , which is a special case of Min-Sum Cycle Packing that asks to find a packing of \( k \) shortest cycles in a graph. We prove this problem to be Fixed-Parameter Tractable (FPT) when parameterized by \( k \) on weighted planar graphs. We also obtain a polynomial kernel for the edge-disjoint variant of the problem on planar graphs. Whether Min-Sum Cycle Packing is FPT on planar graphs, or Shortest Cycle Packing on general graphs, remains open.