MXene-based microwave absorbers have received extensive attention owing to their high electrical conductivity, abundant interfacial polarization sites, and tunable surface terminations. However, the structure–property relationship of MXene composites remains highly nonlinear, and the design of high-efficiency absorbers still relies heavily on trial-and-error experiments. Herein, multidimensional magnetic components, including zero-dimensional (0D) Fe3O4 nanoparticles, one-dimensional (1D) Fe3O4/Co3O4 nanowires, and two-dimensional (2D) Fe3O4-based heterostructures, were rationally integrated with Fe/MXene and Fe/Co/MXene nanosheets to engineer synergistic dielectric and magnetic losses. Comprehensive electromagnetic characterization and loss mechanism analysis reveal that the structural dimensionality strongly impacts impedance matching and attenuation capability. To further enable predictive and data-driven optimization, a machine learning framework was established to correlate the microstructure, component ratio, thickness, and electromagnetic parameters with the microwave absorption performance (e.g., minimum reflection loss (RLmin), effective absorption bandwidth (EAB)). The optimized multidimensional composite achieves an RLmin of −56.4 dB at 10.2 GHz with an EAB of 8.4 GHz (9.6–18.0 GHz) at a thin matching thickness of 1.8 mm. The machine learning model demonstrates excellent accuracy (R2 = 0.947) and enables the inverse design of absorber geometries to target specific operational frequencies. This work provides a generalizable paradigm for the intelligent design of MXene-based microwave absorbers and opens up broader opportunities for the AI-accelerated discovery of advanced electromagnetic functional materials.

Financial time series forecasting poses significant challenges due to the diverse risk profiles and dynamic behaviors of assets such as the S&P 500, NASDAQ, and Bitcoin, especially across different market periods. This study introduces a novel framework, waveletstochastic-chaos informed machine learning, that integrates wavelet transforms, stochastic processes, and chaos theory to improve machine learning prediction accuracy over a decade (2015–2025). The analysis is divided into four distinct periods: All Time, PreCOVID, COVID, and Post-COVID. The aim is to capture the multi-scale patterns, volatility, and complexity inherent in financial data, which will be assessed across various market conditions. The framework outperforms the baseline maximum likelihood models in most scenarios, achieving significant root mean squared errors for the scaled price predictions of S&P 500 (e.g., from 0.0348 to 0.0122 in All Time), NASDAQ (e.g., from 0.0284 to 0.0180 in All Time) and Bitcoin (e.g., from 0.0838 to 0.0288 in All Time) based on 1000 experimental trials. It excels in volatile periods like COVID and for high-risk Bitcoin, though it slightly underperforms in the stable Post-COVID recovery for S&P 500. Wavelet features are found to be critical for accuracy. Additionally, stochastic and chaos-based elements enhance performance in volatile and complex contexts, respectively, as confirmed by ablation studies. This study provides empirical evidence of predictive utility for financial time-series forecasting in assets with different dynamics and market regimes. The results indicate that multi-scale, stochastic, and complexity-based feature representations can improve forecasting performance within the examined datasets, suggesting that the framework may apply to other non-stationary time-series settings, although such extensions remain for future investigation.

Predicting carbon dioxide (CO2) solubility in brine is critical for carbon capture and storage. This study employs the Ant Colony Optimization (ACO) algorithm to enhance the predictive accuracy of four machine learning models: Neural Network (NN), Decision Tree (DT), Support Vector Regression (SVR), and Gradient Boosting Machine (GBM). The models were trained and validated on a mineral compound dataset. Performance was evaluated using the coefficient of determination (R2) and error metrics including RMSE and MAE. The GBM model achieved the highest test accuracy (R2 = 0.986) with low errors (RMSE = 0.0478, MAE = 0.0362), demonstrating superior ability to model complex, non-linear relationships with minimal overfitting. The optimized NN, featuring three layers and fifteen neurons, delivered strong performance (R2 = 0.930) with balanced errors across datasets. The DT model offered excellent interpretability and a strong test score (R2 = 0.912), while the SVR model provided robust generalization (R2 = 0.889). The results indicate that ACO is an effective tool for hyperparameter tuning across diverse model architectures. For maximum accuracy, GBM is recommended, whereas DT is ideal when interpretability is required. The NN presents a strong middle-ground option with competitive accuracy. This comparative framework assists in selecting the optimal model based on specific project priorities of accuracy, transparency, or computational efficiency for geochemical forecasting.

The applications of machine learning (ML) in power electronics are expanding with time, providing effective tools that reduce design complexity and enhance predictive accuracy. In high-power semiconductor devices, such as thyristors and high-power diodes, electrical parameters may directly influence electro-thermal behavior, reliability, and overall device performance. Consequently, accurate prediction and classification of average current are critical to ensure optimal device selection, optimize design, and assess performance. In this article, a comprehensive dataset based on data from industrial thyristors capturing electrical and structural parameters relevant to current handling capability is utilized to classify and predict the average current of devices. Additionally, Shapley additive explanation (SHAP) analysis has been performed, highlighting the importance of crucial parameters and identifying the impact of each parameter on model output. Moreover, several ML models, including artificial neural networks (ANNs), support vector machines (SVMs), ensembles, and Gaussian process regression (GPR) are implemented and then compared to assess their performance. The proposed methodology provides manufacturers and designers with data-driven design tools that enhance reliability assessments and facilitate optimized device selection for high-power applications.

Accurate rainfall classification is essential for Bangladesh, where monsoon variability strongly influences agriculture, water resource management, and disaster preparedness. This study proposes a robust machine learning framework for rainfall intensity classification at the daily temporal scale and nationwide spatial coverage, using over 543,839 daily weather records collected from 35 meteorological stations across several decades from a publicly available national meteorological dataset. The dataset includes rainfall, temperature, humidity, and sunshine duration, which were preprocessed and categorized into four intensity levels: No Rain, Light Rain, Moderate Rain, and Very Heavy Rain. Various models were evaluated, including Random Forest, Decision Trees, Gradient Boosting, K-Nearest Neighbors, Naïve Bayes, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost), along with deep learning architectures such as Artificial Neural Network (ANN), Deep Neural Network (DNN), One-Dimensional Convolutional Neural Network (1D-CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM). Random Forest achieved the highest accuracy (77.37%), while Bi-LSTM performed best among deep learning models (76.97%). To address class imbalance, we adopted class weighting in the final models; SMOTE was explored as an ablation and then excluded due to poorer generalization. Model interpretability using Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) consistently identified humidity and sunshine as the most influential predictors, with SHAP further revealing strong interactions between lagged humidity and temperature. The framework‘s reliable classification of rainfall intensities supports data-driven irrigation scheduling, early flood warnings, and climate-resilient agricultural and disaster management planning in Bangladesh.

Obesity is a major health concern linked to chronic conditions such as diabetes and cardiovascular disease. However, most neurological studies have focused on specific metabolic states, limiting understanding of how brain function changes from fasting to satiety. Furthermore, hypothesis-driven approaches may introduce bias and fail to capture complex neural interactions. This study aimed to identify brain connectivity patterns associated with obesity across different metabolic states using a data-driven approach. Electroencephalography data were collected from 30 women with obesity and 30 women without obesity over a four-hour period encompassing fasting and post-meal states. All subjects were aged 20 to 65 years. Functional connectivity was calculated from source-localized signals, and a machine learning framework incorporating a feature selection method was applied to identify the most discriminative connectivity features between groups. Here we show that six connectivity features classify obesity with 95% accuracy across metabolic states. Reduced connectivity are observed within food-reward processing regions in the obese group, with the dorsal anterior cingulate cortex emerging as a central hub. This pattern reflects a persistent alteration in energy prediction and craving regulation that is independent of metabolic state. These findings demonstrate that disrupted brain connectivity is a fundamental characteristic of obesity. The results highlight the dorsal anterior cingulate cortex as a key region underlying maladaptive reward processing and suggest that targeting this area through neuromodulation therapies may offer a promising intervention for obesity treatment.

Abstract The prediction of building settlement around foundation pit is of vital importance to ensure the safety and stability of urban construction projects. However, current predictions of buildings surrounding foundation pit face numerous challenges. Therefore, this study proposes a framework that integrates Optuna-based hyperparameter optimization, Variational Mode Decomposition (VMD), and machine learning (ML) for accurate and timely settlement prediction in practical engineering scenarios, referred to as the Optuna–VMD–ML framework. Optuna is employed to tune the hyperparameters of both VMD and the ML models; the Optuna-tuned VMD decomposes the settlement time series into mode components, which are then used to train the ML predictors to forecast settlement. By comparing the settlement prediction performance of four different ML models at 12 monitoring points around the Moulding Building in Taizhou, the results show that the Long Short-Term Memory (LSTM) model yields the best prediction accuracy under the proposed framework, with average R2=0.927, MSE=0.013mm2, RMSE=0.107mm, MAE=0.087mm, and MAPE=1.539% across all monitoring points. More importantly, the predicted settlement displacement and velocity are used as core evaluation indicators to construct a comprehensive safety risk assessment framework for buildings around the foundation pit. The proposed method in this paper can effectively realize accurate prediction of settlement around foundation pit buildings, while accurately identifying high-risk areas for structural safety of surrounding buildings. It provides important reference for dynamic monitoring and engineering safety based on settlement prediction.

Background: Ultrasound imaging is an ideal tool for regular carotid plaque screening to identify individuals at high risk of stroke for clinical intervention. However, no existing study leverages multi-modal multi-view ultrasound imaging for AI-enabled auto-classification of carotid plaque vulnerability. This study aims to develop and validate an effective AI model for carotid plaque vulnerability classification through the applications of dual-modal (B-Mode and contrast-enhanced mode) dual-view (longitudinal and cross-sectional) settings to maximize the utility and potential of ultrasound imaging. Methods: Hybrid deep-learning (DL) and machine-learning (ML) methods were employed to balance between model discriminability and interpretability. B-Mode ultrasound (BMUS) and contrast-enhanced ultrasound (CEUS) images from 241 patients were retrospectively analyzed using the proposed hybrid-DL-ML variants. Results: Our findings suggest the hybrid VGG-RF model developed from a dual-modal dual-view setting outperforms those developed from other settings for identifying vulnerable carotid plaques. The VGG-RF model emerged as the best-performing model, achieving an optimal performance with an AUC of 0.908, precision of 0.765, recall of 0.929, specificity of 0.886, and F1 score of 0.839. The inherent interpretability of the VGG-RF model divulged that long-axis views of BMUS and CEUS images were the major contributing features for discriminating vulnerable carotid plaques against their counterparts. Conclusions: The present study underscored the effectiveness of AI models developed from dual-modal dual-view settings of ultrasound images. Notably, the hybrid VGG-RF model was benchmarked as the best-performing model among other studied hybrid DL-ML variants. Further studies on a larger cohort in a prospective setting are warranted to validate the findings of the current study.

Background and Objectives: Uterine fibroids are common benign tumors that frequently require surgical management, particularly myomectomy, in women of reproductive age. Metabolic dysfunction and insulin resistance have been implicated in fibroid biology; however, the clinical relevance of glycated hemoglobin (HbA1c) in predicting myomectomy requirement remains unclear. This study aimed to evaluate the predictive role of HbA1c for myomectomy requirement in women with uterine fibroids using conventional statistical analyses and machine learning-based models under real-world clinical decision-making conditions. Materials and Methods: This study evaluated data from a retrospective multicenter cohort comprising 618 women with a diagnosis of uterine fibroids. Patients were stratified according to myomectomy status (performed vs. not performed). Comparative analyses, univariate and multivariate logistic regression, and machine learning modeling were conducted using demographic, laboratory, hormonal, and fibroid-related variables. A total of 155 machine learning models were trained, and the top 20 models with the highest accuracy were evaluated. Blinded concordance analysis was conducted on 50 independent, anonymized cases evaluated by a gynecologist who was blinded to the study data. Results: Patients undergoing myomectomy (38.5%) had significantly higher HbA1c levels than non-surgical patients (5.57 ± 0.32 vs. 5.03 ± 0.61, p < 0.001). HbA1c showed a strong association with myomectomy requirement in univariate analysis (OR 0.026, 95% CI 0.012–0.055) but lost significance in multivariate models, while ferritin remained independently associated. Machine learning models incorporating HbA1c, ferritin, hormonal, and fibroid parameters achieved accuracies between 0.99 and 1.00. Blinded concordance analysis demonstrated 94% concordance between model predictions and expert clinical judgment. Conclusions: HbA1c is a valuable integrative marker in predicting myomectomy requirement when evaluated within multidimensional machine learning frameworks, although its independent effect is confounded by iron-related parameters. These findings support the use of HbA1c as part of a comprehensive decision-support approach in uterine fibroid management.

Accurate prediction of vehicle CO₂ emissions is challenging due to heterogeneous engine characteristics, nonlinear interactions among fuel, mechanical, and operational parameters, and variable driving conditions. This study proposes a high-performance machine learning framework that combines multi-layer perceptron (MLP) architectures with nature-inspired metaheuristic optimization to model vehicle-induced CO₂ emissions with improved precision and convergence stability. The framework leverages multi-modal engine telemetry-including fuel type, transmission, engine displacement, consumption metrics, and cylinder profiles-alongside advanced feature selection and interpretability techniques such as Recursive Feature Elimination (RFE), SHAP analysis, and Class Activation Mapping (CAM) to identify dominant emission drivers. Two metaheuristic optimizers, Horned Lizard Optimization Algorithm (HLOA) and Giant Armadillo Optimization (GAO), are applied for hyperparameter tuning, with the GAO-enhanced MLP achieving superior predictive performance (R² = 0.9881; RMSE = 6.478). The study highlights the integration of interpretable AI models into vehicle emission prediction, demonstrating their potential to inform low-carbon vehicle design, data-driven urban mobility planning, and environmentally conscious policy-making.

Motor temperature prediction is critical for ensuring the reliability and safe operation of high-power-density electric drives. Since direct measurement of internal temperatures, especially rotor and magnet temperatures, is often impractical, virtual sensing has become an important research direction. This review provides a structured clarification of motor temperature prediction technologies. First, the physical foundations of motor thermal behavior are revisited, emphasizing multi-source loss generation, electro-thermal coupling mechanisms, and the dominant influence of time-varying boundary conditions. Second, existing estimation methodologies are systematically categorized into physics-based thermal models, observer- and identification-based approaches, and data-driven machine learning frameworks. Their mathematical principles, information bottlenecks, computational trade-offs, and deployment constraints are comparatively analyzed. Particular attention is given to hybrid and physics-informed methods, where reduced-order thermal networks, parameter adaptation, and learning-based residual correction are integrated to enhance robustness. Future developments should focus on uncertainty-aware estimation, lifecycle-adaptive modeling, and reliable temperature field inference under sparse sensing conditions.

Resource-stratified machine learning framework for cognitive status classification and mild cognitive impairment to dementia progression prediction.

Prediction of Groundwater Total Nitrogen via an Interpretable Ensemble Machine learning Framework: Implications for Groundwater Diversion Management in Complex Catchments.

Machine learning framework for early detection of polio outbreaks from acute flaccid paralysis surveillance data.

Antimicrobial peptides (AMPs) offer promising alternatives to conventional antibiotics, yet most predictive models fail to account for chemical modifications that influence real-world efficacy. Among these, C-terminal amidation is a widely adopted and effective strategy that improves structural stability, membrane interaction, and protease resistance. In this study, we established an integrated framework for the design and prediction of C-terminal amidated AMPs targeting Escherichia coli. Our approach combined a design-oriented model based on an interpretable Explainable Boosting Machine (EBM), which extracts actionable sequence-level design rules, together with a reliable deployment model, built on a fine-tuned ESM2 deep learning architecture. The resulting tool, CAmidPred, enables both predictive classification and amino acid pattern analysis with outputs examined in relation to published alanine-scanning experiments. Using these models, we identified a pardaxin variant with improved activity against E. coli, demonstrating the practical utility of the dual-model framework in targeted AMP design.

Abstract While the futures–spot price relationship is well established in commodity markets, the transmission of price signals to the retail level remains an " incomplete bridge ,” particularly under varying speculative regimes. Traditional empirical approaches often fail to capture the nonlinear and heterogeneous dynamics of this process, typically providing a single Average Treatment Effect (ATE) that masks the distortions caused by market frictions. This study addresses this gap by developing a novel causal machine learning (CML) framework. Leveraging double machine learning (DML), we isolate the causal link between futures and retail prices by " partialing out ” high-dimensional confounding variables, effectively distinguishing the informational signal from the market " noise ” identified in recent literature. We illustrate this framework using the US frozen concentrated orange juice (FCOJ) market as a functional laboratory for concentrated and volatile ” soft ” commodities. Our results reveal a non-monotonic relationship: while moderate speculation enhances price discovery (CATE $$\approx$$ ≈ 1), both low and high speculative intensity impair signal propagation. Crucially, we find that excessive speculation leads to ” informational decoupling ,” where increased statistical uncertainty in the CATE reflects a coordination failure in firm-level pricing and procurement decisions. These findings challenge the assumption that speculation consistently enhances market efficiency and could provide a robust, scalable, data-driven analytical tool for future supply-chain research and policy discussions on commodity market regulation.

This study examines the factors driving perceived Study Efficiency and Exam Readiness associated with ChatGPT use among STEM students in higher education. Although prior research on generative artificial intelligence (GenAI) has largely focused on adoption and attitudes using descriptive or linear statistical approaches, limited empirical work has explored how students’ interactions with such tools relate to learning-related outcomes. To address this gap, this study applies an interpretable machine learning (ML) framework to identify key predictors of learning gains from ChatGPT use. Data were obtained from a large-scale global survey of STEM students (n = 10,525) across 109 countries and territories, capturing usage patterns, perceived capabilities, satisfaction, and academic outcomes. Two eXtreme Gradient Boosting (XGBoost)-based ML classification models were developed to predict Study Efficiency and Exam Readiness, and SHapley Additive exPlanations (SHAP) were used to interpret feature-level contributions. The models achieved strong predictive performance for the high-gain class, with an accuracy of 0.93 (F1 = 0.96) for Study Efficiency and 0.86 (F1 = 0.92) for Exam Readiness. Results indicate that motivation, personalized learning support, improved access to knowledge, facilitation of study activities, and exam-focused study assistance are key predictors of learning gains. These findings offer empirical and practical insights for educators and policymakers seeking to design effective and pedagogically sound AI-assisted learning environments in STEM education.

Rapid and accurate identification of pollution sources is critical for emergency management but remains challenged by the high computational cost and uncertainty of traditional numerical models. To address this, this study aims to develop a novel hybrid framework that integrates machine learning with numerical modeling for efficient and robust source inversion. A MIKE 21 hydrodynamic-water quality model of the Dongliao River was developed to generate a synthetic dataset to train long short-term memory (LSTM), kernel extreme learning machine, and support vector machine surrogate models. Among them, LSTM achieved superior accuracy and was selected for further integration. For deterministic source identification, a whale optimization algorithm (WOA)-LSTM model was developed, significantly reducing both inversion error and computation time. A probabilistic inversion system was subsequently established by coupling the WOA-LSTM model with a Bayesian framework to characterize posterior probability distributions. Comparative analysis under data noise scenarios revealed that while the deterministic method performed poorly, the probabilistic approach demonstrated remarkable robustness, improving inversion accuracy by over 47%. These findings demonstrate that integrating a physics-informed ML surrogate with Bayesian inference effectively addresses the trade-off between efficiency and uncertainty. This framework offers a powerful tool for intelligent early warning systems, supporting decision-makers in the effective management and mitigation of sudden water pollution incidents.

Background Cardiovascular events (CVEs) are the leading cause of mortality in hemodialysis patients. Current prediction models rely on clinical and biochemical data, but non-invasive alternatives are needed. Inspired by the Traditional Chinese Medicine (TCM) principle that “the heart opens into the tongue,” this study investigated whether quantitative features from tongue images could enhance CVE prediction. Objective To develop and validate a machine learning framework that integrates tongue image features with conventional clinical variables to predict CVEs in hemodialysis patients. Methods In this prospective, multicenter study, 506 maintenance hemodialysis patients were recruited. We extracted 1,354 hand-crafted radiomic features and 8 deep-learning features from standardized tongue images. These were combined with 90 clinical variables. Using a dataset split into training (n=243), validation (n=105), and an independent external test set (n=158), we developed and compared four models (LR, LightGBM, AdaBoost, MLP) under three feature configurations: clinical-only, tongue-only, and a fused model. Results The model using only tongue image features (AdaBoost) significantly outperformed the clinical-only model, achieving an AUC of 0.786 vs. 0.682 on the external test set. The fused model provided a marginal improvement (AUC=0.787). SHAP analysis indicated that both tongue texture features and clinical biomarkers like PT% were key predictors. Decision curve analysis confirmed the clinical utility of the tongue-based and fused models across a range of risk thresholds. Conclusion Tongue image features are potent, non-invasive predictors of CVEs in hemodialysis patients, offering performance superior to conventional clinical variables. This AI-driven approach validates the TCM theory and presents a promising supplementary tool for enhancing risk stratification in nephrology care.

Objective To develop an automated machine learning (AutoML) framework integrating body composition indices—notably Body Fat Rate (BFR)—and clinicopathological features for predicting postoperative complications in gastric cancer patients, addressing limitations of traditional body mass index (BMI) assessment and enhancing clinical translatability. Methods In this retrospective cohort study, 1,023 gastric cancer patients undergoing radical gastrectomy (January 2020–January 2025) were enrolled across two hospitals (716 training, 307 testing). A dual-optimization workflow included: (1) Simultaneous feature selection and hyperparameter tuning via the Improved Hike Optimization Algorithm (IHOA); (2) Class imbalance mitigation using synthetic minority oversampling technique (SMOTE). Model performance was evaluated through accuracy, sensitivity, specificity, F1-score, area under the receiver operating characteristic curve (AUC-ROC), area under the precision-recall curve (AUC-PR), calibration curves, and decision curve analysis (DCA). Feature robustness was validated using least absolute shrinkage and selection operator regression, while SHapley Additive exPlanations (SHAP) interpreted predictor contributions. A MATLAB-based proof-of-concept prototype visualization tool was developed for implementation. Results In independent testing, AutoML maintained robust performance (ROC-AUC = 0.9380, PR-AUC = 0.9262). DCA revealed greater net clinical benefit across risk thresholds (1%–93%) compared to conventional methods, with sustained high-level stability confirming superior generalizability. Calibration curves demonstrated optimal probabilistic prediction (lowest test-set Brier score = 0.111). SHAP analysis identified BFR, visceral fat density (VFD), visceral fat area (VFA), skeletal muscle area (SMA), C-reactive protein (CRP), BMI, Age and lymphadenectomy extent as key predictors. Conclusion The AutoML prediction model developed in this study achieves both high precision and strong interpretability. Its visualized tool effectively overcomes barriers to clinical translation, providing intelligent decision support for early warning and personalized intervention of postoperative complications in gastric cancer.