Search Results (2,757)

Aphids are significant pests of winter wheat, causing damage by feeding on plant sap and reducing crop yield and quality. This study evaluates the potential of hyperspectral remote sensing (350–2500 nm) and machine learning (ML) models for classifying healthy and aphid-infested wheat canopies. Field-based hyperspectral measurements were conducted at three growth stages—T1 (stem elongation–heading), T2 (flowering), and T3 (milky grain development)—with infestation levels categorized according to established economic thresholds (ET) for each growth stage. Spectral data were analyzed using Uniform Manifold Approximation and Projection (UMAP); vegetation indices; and ML classification models, including Logistic Regression (LR), k-Nearest Neighbors (KNNs), Support vector machines (SVMs), Random Forest (RF), and Light Gradient Boosting Machine (LGBM). The classification models achieved high performance, with F1-scores ranging from 0.88 to 0.99, and SVM and RF consistently outperforming other models across all input datasets. The best classification results were obtained at T2 with an F1-score of 0.98, while models trained on the full spectrum dataset showed the highest overall accuracy. Among vegetation indices, the Modified Triangular Vegetation Index, MTVI (r_pb = −0.77 to −0.82), and Triangular Vegetation Index, TVI (r_pb = −0.66 to −0.75), demonstrated the strongest correlations with canopy condition. These findings underscore the utility of canopy spectra and vegetation indices for detecting aphid infestations above ET levels, allowing for a clear classification of wheat fields into “treatment required” and “no treatment required” categories. This approach provides a precise and timely decision making tool for insecticide application, contributing to sustainable pest management by enabling targeted interventions, reducing unnecessary pesticide use, and supporting effective crop protection practices. Full article

Fatigue has always been one of the major causes of structural failure, where repeated loading and unloading cycles reduce the fracture energy of the material, causing it to fail at stresses lower than its monotonic strength. However, predicting fatigue life is a highly challenging task and, in this context, the present study proposes a fundamentally new hybrid physics-informed and data-driven approach. Firstly, an energy-based fatigue model is developed to simulate the behavior of concrete under compressive cyclic fatigue loading. The data generated from these numerical simulations are then utilized to train machine learning (ML) models. The stress–strain curve and S-N curve of concrete under compression, obtained from the energy-based model, are validated against experimental data. For the ML models, two different algorithms are used as follows: k-Nearest Neighbors (KNN) and Deep Neural Networks (DNN), where a total of 1962 data instances generated from numerical simulations are used for the training and testing of the ML models. Furthermore, the performance of the ML models is evaluated for out-of-range inputs, where the DNN model with three hidden layers (a complex model with 128, 64, and 32 neurons) provides the best predictions, with only a 0.6% overall error. Full article

This study presents a novel approach for filament recognition in fused filament fabrication (FFF) processes using a multi-spectral spectroscopy sensor module combined with machine learning techniques. The sensor module measures 18 wavelengths spanning the visible to near-infrared spectra, with a custom-designed shroud to ensure systematic data collection. Filament samples include polylactic acid (PLA), thermoplastic polyurethane (TPU), thermoplastic copolyester (TPC), carbon fibre, acrylonitrile butadiene styrene (ABS), and ABS blended with Carbon fibre. Data are collected using the Triad Spectroscopy module AS7265x (composed of AS72651, AS72652, AS72653 sensor units) positioned at three measurement distances (12 mm, 16 mm, 20 mm) to evaluate recognition performance under varying configurations. Machine learning models, including k-Nearest Neighbors (kNN), Logistic Regression, Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP), are employed with hyperparameter tuning applied to optimise classification accuracy. Results show that the data collected on the AS72651 sensor, paired with the SVM model, achieves the highest accuracy of 98.95% at a 20 mm measurement distance. This work introduces a compact, high-accuracy filament recognition module that can enhance the autonomy of multi-material 3D printing by dynamically identifying and switching between different filaments, optimising printing parameters for each material, and expanding the versatility of additive manufacturing applications. Full article

The working environment at coal mining faces is harsh, leading to high failure rates and significant maintenance issues with roadheaders. This study explores multi-layer dimensionality reduction of vibration signal features in complex environments to enhance the differentiation of different operational states of a roadheader, thereby achieving fault recognition of key components. Concurrently, reducing dimensionality in manifold spaces positively influences operational state differentiation. Therefore, this paper integrates manifold learning to conduct multi-sensor and multi-layer data mining to enhance the differential phenotypes between faults of key components of the roadheader. Initially, we constructed multiple status-reference sample sets for each sensor individually, forming multiple manifolds at different spatial points, and utilizing locality-preserving projections (LPP) to extract low-dimensional manifold features. Further fusion of low-dimensional features from multiple sensors was used to elevate samples, constructing an enhanced spatial pseudo-manifold. Finally, we used LPP to re-reduce the enhanced sensitive feature set from multiple vibration sensors, establishing a dual-layer sensitive feature enhancement learning model. Conducting fault recognition analysis on experimental vibration signals, using k-nearest neighbors (KNN) to classify the enhanced feature set, we achieved a recognition success rate of 98.75% for samples, proving the method’s feasibility in fault recognition under complex loads. Full article

Pipeline filling and emptying are critical hydraulic procedures involving transient two-phase air–water interactions, which can cause pressure surges and structural risks. Traditional Digital Twin models rely on one-dimensional (1D) approaches, which cannot capture air–water interactions. This study integrates Computational Fluid Dynamics (CFD) models into a Digital Twin framework for improved predictive analysis. A CFD-based Digital Twin is developed and validated using real-time pressure measurements, incorporating 2D and 3D CFD models, mesh sensitivity analysis, and calibration procedures. Key contributions include a CFD-driven Digital Twin for real-time monitoring and machine learning (ML) techniques to optimise pressure surges. ML models trained with experimental and CFD data reduce reliance on computationally expensive CFD simulations. Among the 31 algorithms tested, decision trees, efficient linear models, and ensemble classifiers achieved 100% accuracy for filling processes, while k-Nearest Neighbours (KNN) provided 97.2% accuracy for emptying processes. These models effectively predict hazardous pressure peaks and vacuum conditions, confirming their reliability in optimising pipeline operations while significantly reducing computational time. Full article

Wildfires are increasingly frequent and severe, posing substantial risks to ecosystems, communities, and infrastructure. Accurately mapping wildfire severity (WSM) using remote sensing and machine learning (ML) is critical for evaluating damages, informing recovery efforts, and guiding long-term mitigation strategies. Stacking ensemble ML (SEML) enhances predictive accuracy and robustness by combining multiple diverse models into a single meta-learned predictor. This approach leverages the complementary strengths of individual base learners while reducing variance, ultimately improving model reliability. This study aims to optimize a SEML framework to (1) identify the most effective ML models for use as base learners and meta-learners, and (2) determine the optimal number of base models needed for robust and accurate wildfire severity predictions. The study utilizes six ML models—Random Forests (RF), Support Vector Machines (SVM), k-Nearest Neighbors (KNN), Linear Regression (LR), Adaptive Boosting (AB), and Multilayer Perceptron (MLP)—to construct an SEML. To quantify wildfire impacts, we extracted 118 spectral indices from post-fire Landsat-8 data and incorporated four additional predictors (land cover, elevation, slope, and aspect). A dataset of 911 CBI observations from 18 wildfire events was used for training, and models were validated through cross-validation and bootstrapping to ensure robustness. To address multicollinearity and reduce computational complexity, we applied Linear Discriminant Analysis (LDA) and condensed the dataset into three primary components. Our results indicated that simpler models, notably LR and KNN, performed well as meta-learners, with LR achieving the highest predictive accuracy. Moreover, using only two base learners (RF and SVM) was sufficient to realize optimal SEML performance, with an overall accuracy and precision of 0.661, recall of 0.662, and F1-score of 0.656. These findings demonstrate that SEML can enhance wildfire severity mapping by improving prediction accuracy and supporting more informed resource allocation and management decisions. Future research should explore additional meta-learning approaches and incorporate emerging remote sensing data sources such as hyperspectral and LiDAR. Full article

Transfer learning (TL) has been employed in electroencephalogram (EEG)-based brain–computer interfaces (BCIs) to enhance performance for cross-session and cross-subject EEG classification. However, domain shifts coupled with a low signal-to-noise ratio between EEG recordings have been demonstrated to contribute to significant variations in EEG neural dynamics from session to session and subject to subject. Critical factors—such as mental fatigue, concentration, and physiological and non-physiological artifacts—can constitute the immense domain shifts seen between EEG recordings, leading to massive inter-subject variations. Consequently, such variations increase the distribution shifts across the source and target domains, in turn weakening the discriminative knowledge of classes and resulting in poor cross-subject transfer performance. In this paper, domain adaptation algorithms, including two machine learning (ML) algorithms, are contrasted based on the single-source-to-single-target (STS) and multi-source-to-single-target (MTS) transfer paradigms, mainly to mitigate the challenge of immense inter-subject variations in EEG neural dynamics that lead to poor classification performance. Afterward, we evaluate the effect of the STS and MTS transfer paradigms on cross-subject transfer performance utilizing three EEG datasets. In this case, to evaluate the effect of STS and MTS transfer schemes on classification performance, domain adaptation algorithms (DAA)—including ML algorithms implemented through a traditional BCI—are compared, namely, manifold embedded knowledge transfer (MEKT), multi-source manifold feature transfer learning (MMFT), k-nearest neighbor (K-NN), and Naïve Bayes (NB). The experimental results illustrated that compared to traditional ML methods, DAA can significantly reduce immense variations in EEG characteristics, in turn resulting in superior cross-subject transfer performance. Notably, superior classification accuracies (CAs) were noted when MMFT was applied, with mean CAs of 89% and 83% recorded, while MEKT recorded mean CAs of 87% and 76% under the STS and MTS transfer paradigms, respectively. Full article

This study investigates the use of educational data mining (EDM) techniques to predict student performance and enhance learning outcomes in higher education. Leveraging data from Moodle, a widely used learning management system (LMS), we analyzed 450 students’ academic records spanning nine semesters. Five machine learning algorithms—k-nearest neighbors, random forest, logistic regression, decision trees, and neural networks—were applied to identify correlations between courses and predict grades. The results indicated that courses with strong correlations (+0.3 and above) significantly enhanced predictive accuracy, particularly in binary classification tasks. kNN and neural networks emerged as the most robust models, achieving F1 scores exceeding 0.8. These findings underscore the potential of EDM to optimize instructional strategies and support personalized learning pathways. This study offers insights into the effective application of data-driven approaches to improve educational outcomes and foster student success. Full article

Natural gas, as a sustainable and cleaner energy source, still holds a crucial position in the energy transition stage. In shale gas exploration, total organic carbon (TOC) content plays a crucial role, with log data proving beneficial in predicting total organic carbon content in shale reservoirs. However, in complex coal-bearing layers like the marine–continental transitional Shanxi Formation, traditional prediction methods exhibit significant errors. Therefore, this study proposes an advanced, cost- and time-saving deep learning approach to predict TOC in marine–continental transitional shale. Five well log records from the study area were used to evaluate five machine learning models: K-Nearest Neighbors (KNNs), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGB), and Deep Neural Network (DNN). The predictive results were compared with conventional methods for accurate TOC predictions. Through K-fold cross-validation, the ML models showed superior accuracy over traditional models, with the DNN model displaying the lowest root mean square error (RMSE) and mean absolute error (MAE). To enhance prediction accuracy, δR was integrated as a new parameter into the ML models. Comparative analysis revealed that the improved DNN-R model reduced MAE and RMSE by 57.1% and 70.6%, respectively, on the training set, and by 59.5% and 72.5%, respectively, on the test set, compared to the original DNN model. The Williams plot and permutation importance confirmed the reliability and effectiveness of the enhanced DNN-R model. The results indicate the potential of machine learning technology as a valuable tool for predicting crucial parameters, especially in marine–continental transitional shale reservoirs lacking sufficient core samples and relying solely on basic well-logging data, signifying its importance for effective shale gas assessment and development. Full article

Red jujube is a nutritious food, known as the “king of all fruits”. The quality of Chinese red jujube is closely associated with its place of origin. To classify Chinese red jujube more correctly, based on the combination of adaptive boosting (Adaboost) and common vectors linear discriminant analysis (CLDA), Adaboost-CLDA was proposed to classify the near-infrared (NIR) spectra of red jujube samples. In the study, the NIR-M-R2 spectrometer was employed to scan red jujube from four different origins to acquire their NIR spectra. Savitzky–Golay filtering was used to preprocess the spectra. CLDA can effectively address the “small sample size” problem, and Adaboost-CLDA can achieve an extremely high classification accuracy rate; thus, Adaboost-CLDA was performed for feature extraction from the NIR spectra. Finally, K-nearest neighbor (KNN) and Bayes served as the classifiers for the identification of red jujube samples. Experiments indicated that Adaboost-CLDA achieved the highest identification accuracy in this identification system for red jujube compared with other feature extraction algorithms. This demonstrates that the combination of Adaboost-CLDA and NIR spectroscopy significantly enhances the classification accuracy, providing an effective method for identifying the geographical origin of Chinese red jujube. Full article

Laminated composites display exceptional weight-saving abilities that make them suited to advanced applications in aerospace, automobile, civil, and marine industries. However, the orthotropic nature of laminated composites means that they possess several damage modes that can lead to catastrophic failure. Therefore, machine learning-based Structural Health Monitoring (SHM) techniques have been used for damage detection. While Lamb waves have shown significant potential in the SHM of laminated composites, most of these techniques are focused on imaging-based methods and are limited to damage detection. Therefore, this study aims to localize the damage in laminated composites without the use of imaging methods, thus improving the computational efficiency of the proposed approach. Moreover, the machine learning models are generally black-box in nature, with no transparency of the reason for their decision making. Thus, this study also proposes the use of Shapley Additive Explanations (SHAP) to identify the important feature to localize the damage in laminated composites. The proposed approach is validated by the experimental simulation of the damage at nine different locations of a composite laminate. Multi-feature extraction is carried out by first applying the Hilbert transform on the envelope signal followed by statistical feature analysis. This study compares raw signal features, Hilbert transform features, and multi-feature extraction from the Hilbert transform to demonstrate the effectiveness of the proposed approach. The results demonstrate the effectiveness of an explainable K-Nearest Neighbor (KNN) model in locating the damage, with an $R^2$ value of 0.96, a Mean Square Error (MSE) value of 10.29, and a Mean Absolute Error (MAE) value of 0.5. Full article

Background: Cardiovascular disease (CVD) is a prominent determinant of mortality, accounting for 17 million lives lost across the globe each year. This underscores its severity as a critical health issue. Extensive research has been undertaken to refine the forecasting of CVD in patients using various supervised, unsupervised, and deep learning approaches. Methods: This study presents HeartEnsembleNet, a novel hybrid ensemble learning model that integrates multiple machine learning (ML) classifiers for CVD risk assessment. The model is evaluated against six classical ML classifiers, including support vector machine (SVM), gradient boosting (GB), decision tree (DT), logistic regression (LR), k-nearest neighbor (KNN), and random forest (RF). Additionally, we compare HeartEnsembleNet with Hybrid Random Forest Linear Models (HRFLM) and ensemble techniques including stacking and voting. Results: Employing a dataset of 70,000 cardiac patients with 12 clinical attributes, our proposed model achieves a notable accuracy of 92.95% and a precision of 93.08%. Conclusions: These results highlight the effectiveness of hybrid ensemble learning in enhancing CVD risk prediction, offering a promising framework for clinical decision support. Full article

Plastic pollution is becoming an increasingly serious threat to the natural environment. Macroplastics, primarily polyethylene films, pose significant ecological and economic risks, particularly in the agricultural sector. Effective monitoring of their presence is necessary to evaluate the effectiveness of mitigation measures. Conventional techniques for identifying environmental contaminants, based on field studies, are often time-consuming and limited in scope. In response to these challenges, a study was conducted with the primary aim of utilizing unmanned aerial vehicles (UAVs), multispectral cameras, and classification tools to monitor macroplastic pollution. The model object for the study was an industrial compost pile. The performance of four object-oriented classifiers—Random Forest, k-Nearest Neighbor (k-NN), Maximum Likelihood, and Minimum Distance—was evaluated to effectively identify waste contamination. The best results were achieved with the k-NN classifier, which recorded a Matthews Correlation Coefficient (MCC) of 0.641 and an accuracy (ACC) of 0.891. The applied classifier identified a total 37.35% of the studied compost pile’s surface as contamination of plastic. The results of the study show that UAV technology, combined with multispectral imaging, can serve as an effective and relatively cost-efficient tool for monitoring macroplastic pollution in the environment. Full article

Accurate identification of growth and development stages is critical for orthodontic diagnosis, treatment planning, and post-treatment retention. While hand–wrist radiographs are the traditional gold standard, the associated radiation exposure necessitates alternative imaging methods. Lateral cephalometric radiographs, particularly the maturation stages of the second, third, and fourth cervical vertebrae (C2, C3, and C4), have emerged as a promising alternative. However, the nonlinear dynamics of these images pose significant challenges for reliable detection. This study presents a novel approach that integrates chaotic functional connectivity (FC) matrices and fractal dimension analysis to address these challenges. The fractal dimensions of C2, C3, and C4 vertebrae were calculated from 945 lateral cephalometric radiographs using three methods: fast Fourier transform (FFT), box counting, and a pre-processed FFT variant. These results were used to construct chaotic FC matrices based on correlations between the calculated fractal dimensions. To effectively model the nonlinear dynamics, chaotic maps were generated, representing a significant advance over traditional methods. Feature selection was performed using a wrapper-based approach combining k-nearest neighbors (kNN) and the Puma optimization algorithm, which efficiently handles the chaotic and computationally complex nature of cervical vertebrae images. This selection minimized the number of features while maintaining high classification performance. The resulting AI-driven model was validated with 10-fold cross-validation and demonstrated high accuracy in identifying growth stages. Our results highlight the effectiveness of integrating chaotic FC matrices and AI in orthodontic practice. The proposed model, with its low computational complexity, successfully handles the nonlinear dynamics in C2, C3, and C4 vertebral images, enabling accurate detection of growth and developmental stages. This work represents a significant step in the detection of growth and development stages and provides a practical and effective solution for future orthodontic diagnosis. Full article

Wind energy is a crucial renewable resource for sustainable power generation; however, challenges such as high initial investment costs and difficulties in identifying efficient locations hinder its widespread adoption. Accurate wind energy forecasting is essential for energy planning, trading, and grid optimization. This study presents short-term, medium-term, and long-term –wind power forecasts for the Söke–Çatalbük Wind Power Plant in Aydın, Turkey, using meteorological data and production records from 2018 to 2022. Five machine learning algorithms were employed—Artificial Neural Network (ANN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), K-Nearest Neighbors Regression (KNN), and Multi-Layer Perceptron (MLP ANN)—utilizing both MinMax and Standard Scaling methods. Prediction performance was evaluated using Mean Absolute Error (MAE), Coefficient of Determination (R²), and Root Mean Square Error (RMSE) metrics. The results indicate that Min-Max Scaling improved short-term predictions with KNN, while XGBoost and Random Forest provided more stable and accurate forecasts in medium- and long-term predictions. Additionally, Standard Scaling significantly enhanced MLP ANN’s performance in medium-term forecasting. These findings provide practical insights for optimizing wind energy forecasting models, which can improve energy trading strategies, enhance grid stability, and support informed decision making in renewable energy investments. The results are particularly valuable for energy planners and policymakers seeking to maximize the efficiency of wind power plants and facilitate the integration of renewable energy sources into national grids more effectively. Full article