Search Results (9,530)

Biomimetic sensors offer a rapid route for estimating active constituent contents in Chinese medicinal materials, but their practical use at procurement sites and production workshops is constrained by small and skewed datasets, repeated task-specific model configuration, and the limited operational traceability of many nonlinear models. This study proposes the Traditional Chinese Medicine Unified Sensor-based Prediction (TCM-USP) framework for traceable preliminary screening rather than confirmatory laboratory quantification. The same modeling workflow was applied across different medicinal material–sensor combinations and integrates controlled symbolic feature construction with density-aware robust partial least squares modeling. The framework was evaluated on six prediction tasks involving five medicinal materials and three sensor types. TCM-USP achieved higher $R^2$ values than raw-feature PLS in all six tasks and achieved the highest $R^2$ among the evaluated models in four tasks. Although SVR or GPR achieved higher $R^2$ values in the remaining two tasks, TCM-USP generated compact prediction formulas directly expressed in terms of the original sensor readings, enabling independent calculation, audit, and rapid batch-level decision support. These results support the feasibility of TCM-USP for traceable preliminary screening across the investigated small-sample and low-dimensional sensor tasks, while pharmacopoeial methods remain necessary for confirmatory quantification. Full article

Credit risk assessment is a core component of financial decision-making. This study develops an explainable machine learning framework for modeling loan approval decisions on heterogeneous tabular data, centered on a Cross-Attentional Tabular Transformer that applies bidirectional cross-attention between numerical and categorical feature groups. The prediction target is historical loan-approval status, treated as a proxy for, not a direct measure of, borrower default risk; a supplementary validation on a dataset with an authentic default label is also reported. Class imbalance is addressed through focal loss, and post hoc interpretability is provided through SHAP analysis. Three classifiers, Random Forest, Gradient Boosting, and the proposed transformer, are evaluated on a 5000-sample credit dataset using accuracy, precision, recall, F1-score, ROC-AUC, and average precision. Gradient Boosting achieves the best performance (accuracy 0.9640, F1-score 0.9189), with Random Forest comparable; the proposed transformer reaches 0.9530 accuracy and 0.8949 F1, without surpassing the ensembles and at substantially higher computational cost. A five-split robustness comparison additionally evaluates XGBoost, LightGBM, CatBoost, and calibrated logistic regression: all three Gradient-Boosting variants and both classical ensembles exceed the transformer’s performance on every metric, while calibrated logistic regression does not. The evaluated baseline set excludes deep tabular architectures such as TabNet, FT-Transformer, SAINT, and TabPFN-style methods. Across the three primary classifiers, SHAP identifies credit score, employment status, and income as the dominant features, consistent with domain expectations. The results characterize the observed performance–efficiency trade-off between ensemble methods and attention-based tabular learning under the evaluated data conditions. Full article

Traditional approaches for monitoring aquatic organic carbon, such as satellite remote sensing and automated underwater sensors, are often constrained by limited temporal resolution, data gaps under cloudy conditions, maintenance requirements, and cost-effectiveness. To overcome these limitations, we developed and field-demonstrated a ground-based hyperspectral remote sensing system (GHRSS) for continuous, high-frequency monitoring of dissolved organic carbon (DOC) and particulate organic carbon (POC). The system is based on the above-water method and integrates three miniature hyperspectral spectrometers to measure water-surface radiance, sky radiance, and downwelling irradiance for deriving hyperspectral remote sensing reflectance ($R_{rs}$). The spectrometers cover 400–900 nm with a spectral resolution of 1 nm and support a minimum sampling interval of 10 s. The GHRSS also integrates solar power supply, 4G communication, and a microcomputer, enabling autonomous long-term deployment and wireless data transmission. Based on the GHRSS, retrieval models for DOC and POC were developed and validated using 90 paired in situ measurements collected from the Cao’e River. Empirical and machine learning methods were applied to retrieve DOC and POC from the measured $R_{rs}$ data. The empirical models showed limited retrieval performance, whereas partial least squares regression (PLSR) and support vector regression (SVR) substantially improved model accuracy. Among all models, SVR achieved the best performance on the independent test set, with $R^2=0.979$, RMSE = 0.031 mg/L, and MAE = 0.024 mg/L for DOC and $R^2=0.960$, RMSE = 0.152 mg/L, and MAE = 0.066 mg/L for POC. Using the optimal SVR models, minute-scale time series of DOC and POC were reconstructed from the GHRSS observations. The results revealed pronounced sub-daily variability in both parameters, with DOC varying relatively smoothly, whereas POC exhibited stronger short-term fluctuations and more rapid responses to hydrodynamic changes. These findings demonstrate that the GHRSS, combined with machine learning models, provides an effective and practical approach for continuous, high-frequency monitoring of riverine organic carbon dynamics. Full article

Nanoplastics (NPs; <1000 nm) are persistent soil contaminants that suppress extracellular enzyme activity, the biochemical engine of terrestrial nutrient cycling. Despite a rapidly expanding primary literature, no comprehensive meta-analysis has systematically integrated quantitative effect-size synthesis with interpretable machine learning (ML) approaches to identify and rank the physicochemical drivers of NP-induced soil enzymatic toxicity. Following the PRISMA 2020 statement, we systematically searched four databases (Web of Science, Scopus, PubMed, Google Scholar) from database inception through December 2024 and extracted 413 effect sizes from 113 peer-reviewed studies. Hedges’ g was estimated using three-level random-effects models with restricted maximum likelihood (REML) estimation implemented in the metafor package. Three supervised ML algorithms—random forest (RF), gradient boosting machines (GBMs), and support vector regression (SVR)—were trained using 18 study-level predictors derived from the complete meta-analytic dataset, and SHapley Additive exPlanations (SHAP) were applied to quantify and rank the relative importance of individual predictors. The overall meta-analysis demonstrated a significant inhibitory effect of NPs on soil enzyme activity (Hedges’ g = −0.94; 95% CI: −1.14 to −0.73; k = 413; I² = 78.4%; τ² = 0.412). Among the evaluated enzymes, dehydrogenase activity exhibited the greatest inhibition (g = −1.12), whereas polystyrene nanoplastics produced the strongest adverse effects (g = −1.15). Particles smaller than 100 nm caused approximately 2.6-fold greater inhibition than particles larger than 500 nm, and dose–response meta-regression identified a nonlinear increase in toxicity at concentrations exceeding 200 mg kg⁻¹. The RF model demonstrated the highest predictive performance, explaining 73% of the variance in an independent testing dataset (R² = 0.73; test set n = 83). SHAP analysis identified particle diameter as the most influential predictor, revealing an approximate critical threshold of 150 nm, below which inhibitory effects increased markedly. Higher soil organic carbon concentrations partially mitigated enzymatic inhibition, likely through competitive adsorption and reduced nanoplastic bioavailability. Overall, our findings demonstrate that NP-induced inhibition of soil enzymatic activity is widespread and primarily governed by particle size, exposure concentration, and soil properties. The identified 150 nm threshold should be interpreted as a data-driven hypothesis requiring further validation under environmentally realistic exposure scenarios rather than as a universal regulatory limit. Nevertheless, the integration of three-level meta-analysis with interpretable machine learning (SHAP) provides a robust and reproducible framework for identifying key toxicity drivers and supports future ecological risk assessment and evidence-based regulatory decision-making for nanoplastics. Full article

Precipitation records from meteorological stations frequently contain gaps caused by sensor, power, or transmission failures, creating uncertainty in hydrological, agricultural, and water-resources applications. This study compared two simple baselines (station-specific monthly climatological mean and temporal linear interpolation), deterministic spatial interpolation, direct reanalysis-based replacement, and machine-learning methods for daily precipitation imputation. Daily precipitation from 14 stations in Eastern and Southeastern Türkiye during 1985–2014 was evaluated using an independent final-test set formed by stratified random masking of 15% of complete observations; the remaining 85% was used for calibration, SHapley Additive exPlanations (SHAP) analysis, cross-validation, and hyperparameter optimization. ERA5-Land variables were transferred to the stations, precipitation was calibrated by Empirical Quantile Mapping, and leakage-controlled Kriging estimates were incorporated as predictors in XGBoost, LightGBM, Random Forest, Support Vector Regression, and Multilayer Perceptron models. The station-month climatological mean (RMSE = 5.4820 mm; NSE = 0.0527) and temporal linear interpolation (RMSE = 5.7059 mm; NSE = −0.0262) performed substantially worse than optimized Kriging and IDW. The full-hybrid LightGBM model achieved the best performance (RMSE = 3.2001 mm; MAE = 0.9814 mm; Pearson r = 0.8317; NSE = 0.6772), whereas direct ERA5-EQM replacement was less accurate (RMSE = 5.2252 mm; NSE = 0.1394). Combining local observations, spatial information, and ERA5-Land covariates therefore improved daily precipitation imputation in the study region. Full article

Accurate tool wear monitoring can effectively improve machining quality and reduce tool costs. In this paper, tool wear monitoring was studied using multi-signal fusion based on an integrated machine learning model. Firstly, tool holder strain, acceleration, and AE signals are selected as tool wear monitoring signals based on different types of physical quantities and acceptable installation convenience. Tool wear experiments are conducted to synchronously acquire these signals. After the signal denoising process, 102 features from these signals are extracted, which include time domain, frequency domain, and wavelet packet time-frequency domain features. Then, 15 key features are selected using the minimum redundancy maximum relevance (mRMR) method to realize multi-signal fusion at the feature level. Subsequently, an integrated machine learning model is proposed for tool wear monitoring. Three complementary models, extra trees, random forest, and ridge regression, are selected to construct the integrated model. The results indicate that this strategy achieves a tool wear state classification accuracy of 96.77%, exhibiting higher accuracy than single models. Full article

Rapid growth in connected devices, artificial intelligence applications, and the Internet of Things (IoT) is driving demand for ultra-high data rates, low latency, and energy-efficient communication infrastructure. Wireless optical communication (WOC), including free-space optical (FSO), is recognized as a disruptive technology for 6G and future-generation networks. However, atmospheric visibility critically affects WOC/FSO link availability, capacity, and reliability. This study proposes a machine learning (ML)-based low-visibility classification model that integrates Meteorological Aerodrome Reports (METARs) with microwave-link received-signal (Rx) features. Visibility below 6000 m is predicted at the 1 h, 3 h, and 6 h horizons using 18 months of data from Suvarnabhumi Airport (VTBS) and Don Mueang Airport (VTBD) in Bangkok, Thailand. Four ML algorithms, namely logistic regression, random forest, extreme gradient boosting, and light gradient boosting machine (LGBM), are evaluated against persistence and Terminal Aerodrome Forecast (TAF) baselines. In a 100-round block-bootstrap evaluation, LGBM with METAR-Rx achieved the highest mean F1 scores at the 1 h and 3 h horizons, outperforming TAF by 28 and 20 percentage points at the 1 h horizon for VTBS and VTBD, respectively. SHAP and ablation analyses suggested that current visibility is the dominant predictor, while Rx features provide complementary information and improve F1 performance by approximately 1–4 percentage points. Seasonal analysis shows stronger cool-season performance, while rainy-season prediction remains challenging. Adding visibility-trend features further improves performance, with the best combined model achieving 1 h F1 scores of 0.7253 for VTBS and 0.6495 for VTBD. These findings indicate that integrating the METAR-Rx feature set can support short-term low-visibility classification. Full article

Background: Patients with stage I rectal cancer generally have favorable outcomes after curative surgery, but prognosis is not completely homogeneous. This study explored the prognostic association of preoperative inflammation–nutrition indices with overall survival and developed an interpretable internally validated machine learning survival model. Methods: We retrospectively included 475 patients with primary pathological stage I rectal adenocarcinoma who underwent curative-intent radical surgery at Sichuan University West China Hospital between 2018 and 2021. Patients downstaged to ypStage I after neoadjuvant therapy or treated by local transanal excision without lymph node dissection were excluded. Candidate predictors included age, sex, carcinoembryonic antigen, and routinely available preoperative inflammation–nutrition indices. LASSO-Cox regression was used for feature selection. Six survival models were developed and evaluated using 1000 bootstrap resamples with out-of-bag internal validation. Model performance was assessed at 60 months using time-dependent AUC, C-index, Brier score, calibration, and decision curve analysis. SHAP analysis was used for model interpretation. Results: During a median follow-up of 68 months, 30 deaths occurred. LASSO-Cox regression identified five predictors: age, lymphocyte-to-white blood cell ratio, fibrinogen-to-lymphocyte ratio, albumin-to-alkaline phosphatase ratio, and neutrophil-to-HDL cholesterol ratio. In bootstrap out-of-bag internal validation, XGBoost-Cox achieved the highest, although only marginally higher, discriminative performance among the evaluated models, with a 60-month time-dependent AUC of 0.783, a C-index of 0.774, and a Brier score of 0.0507. The calibration intercept and slope of XGBoost-Cox were 0.519 and 1.070, respectively. SHAP analysis identified age as the most influential predictor, followed by the selected inflammation–nutrition indices. Decision curve analysis suggested potential clinical utility within threshold probabilities from 1% to 20%, although this finding remains exploratory. Conclusions: Preoperative inflammation–nutrition indices may contribute to overall survival prognostic stratification in primary pathological stage I rectal cancer. External validation in larger multicenter cohorts is required before clinical application. Full article

To overcome the challenges in conventional frame structure design, namely the difficulty in balancing lightweight design and performance enhancement, the low efficiency of finite element (FE) simulation, and the tedious process of multivariable preliminary screening, an efficient optimization framework for frame structures that integrates multi-objective size optimization, machine learning-based performance prediction, and feature correlation analysis is proposed. First, for the steel–aluminum hybrid frame (with the main load-bearing components made of 6005A aluminum alloy and the critical load-bearing supports and joints made of Q345 low-alloy high-strength steel), a trade-off solution is obtained through multi-objective size optimization. Verified by FE simulation, this solution reduces the frame mass by 6.37% and increases the torsional stiffness by 10.47% compared with the initial design. At the same time, the modal performance, structural strength, and deformation control capability are all significantly improved, achieving a precise balance between lightweighting and stiffness enhancement. Second, a linear regression prediction model is constructed to achieve high-accuracy predictions. The average prediction error for torsional stiffness is only 2%, and the maximum prediction error for the seventh-order modal frequency is less than 1%. The prediction time for a single sample is less than one second, which is more than 1000 times faster than conventional FE simulation, thus efficiently replacing time-consuming simulation analyses. Finally, feature correlation analysis is adopted as an alternative to traditional sensitivity analysis. The core variables identified by this method are highly consistent with those obtained from Hypermesh sensitivity analysis, enabling rapid multivariable screening without additional simulations and greatly improving the efficiency of the preliminary analysis phase. The proposed optimization framework achieves a favorable combination of optimization effectiveness, prediction accuracy, and design efficiency. It not only provides a feasible engineering solution for the lightweight design of frame structures but also serves as a technical reference for the efficient optimization of similar complex structures, demonstrating significant engineering application value. Full article

This study employs machine learning methods to estimate possible source regions from Andean lead isotopes, a novel approach for tracing the provenance of metal artifacts in this region. Two Random Forest coordinate models were trained to approximate geographic provenance. The latitude model uses ²⁰⁶Pb/²⁰⁴Pb, ²⁰⁷Pb/²⁰⁴Pb, and ²⁰⁸Pb/²⁰⁴Pb isotope predictors, plus three derived ratios. The longitude model uses the same predictors together with the latitude predicted by the first model, allowing the regression to incorporate broad spatial coherence in the Andean ore-lead system. The Random Forest models reached, on validation, 3.50° and 1.94° MAE for latitude and longitude, respectively. These moderate values support the use of the framework as a screening tool that generates a ranked shortlist of nearest isotopic matches, rather than as a source of single definitive coordinates. Additionally, K-means clustering and Euclidean distance analysis were used to link artifact isotope compositions to known sources. The models’ limitations and scope were documented to ensure appropriate use and interpretation. Full article

Background: Polypharmacy is associated with increased adverse drug event (ADE) risk in older adults, but accurate risk stratification remains challenging. This study aimed to develop and evaluate explainable machine learning (ML) models for predicting ADEs in older adults with polypharmacy. Methods: This single-center retrospective cohort study included adults aged ≥65 years who received outpatient care at Daegu Catholic University Medical Center between January 2016 and December 2025. Logistic regression, random forest, and Light Gradient-Boosting Machine (LightGBM) models were developed using demographic, comorbidity, medication, and laboratory variables. Model performance was evaluated using discrimination, calibration, and classification metrics. SHapley Additive exPlanations (SHAP) analyses were performed to improve model interpretability. Results: A total of 7505 older adults were included, including 366 patients who developed ADEs within 90 days. In the independent test set, random forest demonstrated favorable overall classification performance, achieving the highest accuracy (0.810), specificity (0.828), and F1-score (0.193), whereas logistic regression showed the highest AUROC (0.705) and sensitivity (0.589). SHAP analyses identified medication count, sodium level, diabetes mellitus, and comorbidity burden as major contributors to ADE risk prediction. Conclusions: Explainable ML models demonstrated moderate but clinically meaningful performance for predicting ADEs in older adults with polypharmacy. These findings suggest that explainable ML approaches may support clinically interpretable medication safety risk stratification in real-world clinical practice. Full article

Reference evapotranspiration (ET_o) is essential for irrigation scheduling, water resources management, and climate-related applications, but the FAO Penman–Monteith (FAO-PM) method is often constrained by limited meteorological data availability. This study evaluated four machine learning (ML) algorithms, Kernel Approximation Regression (KAR), Multilayer Perceptron (MLP), Extreme Gradient Boosting (XGB), and Random Forest (RF), for reproducing FAO-PM ET_o under different levels of geographical and meteorological information availability in the Czech Republic. Daily observations from 59 meteorological stations (1980–2024) were used to develop eight input scenarios. Model performance was evaluated using a station-wise chronological train–test framework and station-based analyses. The results showed that predictor availability had a greater influence on model performance than model selection. The geographical-information scenario produced the lowest performance, whereas substantial improvements were achieved when meteorological variables were incorporated. Among the single-variable meteorological scenarios, relative humidity provided the greatest improvement in agreement with the FAO-PM ET_o benchmark. Across all input scenarios and ML algorithms, testing performance ranged from R² = 0.683 to 0.998 and RMSE = 0.076 to 0.939 mm d⁻¹, indicating progressively improved agreement with FAO-PM ET_o as additional meteorological information became available. The reduced-input scenarios therefore provide a practical approach for approximating FAO-PM ET_o at stations represented during model development when some meteorological inputs are unavailable. Full article

This study investigated the manufacturing and impact of critical input parameters in fused filament fabrication (FFF) on the ultimate tensile strength (UTS) of tailor-made recycled PLA/wood bio-composite specimens. As technology advances rapidly, several wood-based polymer composites have emerged as promising materials for wood-based interior applications. In the current work, recycled PLA material is combined with wood powders to form composite 3D printing filaments. The solution casting method is used to recycle the PLA and a single-screw extruder is used to fabricate the composite filament. 3D printing parameters play a major role in enhancing the characteristics of the wood-based polymers. This study considers the printing temperature (PT), layer height (LH), and printing speed (PS) as input parameters at five levels. Taguchi Design of Experiments (L25 orthogonal array) was employed to minimize experimental runs, followed by ANOVA analysis to find influencing factors. The results demonstrated that layer height (83.91% contribution) is the most critical parameter, with 0.1 mm identified as the optimal amount for achieving the maximum UTS response, while printing temperature (2.47%) had a moderate effect and printing speed (2.28%) showed negligible influence. In the present work, the tensile properties of the composite filament were predicted using machine learning methodologies, including random forest (RF), support vector regressor (SVR), Gradient Boosting Regression (GBR), Extreme Gradient Boosting (XG Boost), and Adaptive Boosting (Adaboost). The results indicate that support vector regressor (SVR) outperformed all other models in terms of generalization, as it generated the lowest test errors (mean squared error (MSE) = 0.0169, mean absolute error (MAE) = 0.0953, mean squared logarithmic error (MSLE) = 0.0065 and mean absolute percentage error (MAPE) = 0.2246) and the highest predictive power (coefficient of determination (R²) = 0.8679). Full article

Background: Caregiver burden represents a multidimensional syndrome influenced by patient-related, relational, and contextual factors in neurological disorders. Although stroke, Parkinson’s disease (PD), and Alzheimer’s disease (AD) differ in clinical trajectory, comparative analyses of caregiver risk profiles across these conditions remain limited. Objective: This study aimed to identify sociodemographic, cognitive, and dyadic factors associated with caregiver burden in a clinical cohort, and to investigate their value for caregiver risk stratification using supervised machine learning models. Methods: In this monocentric observational cohort study, 113 patient–caregiver dyads (79 stroke, 19 PD, 15 AD) were consecutively enrolled in a neurorehabilitation setting. Patients underwent cognitive assessment with the Montreal Cognitive Assessment (MoCA), while caregivers completed the Caregiver Burden Inventory (CBI), which was considered the primary outcome measure. Caregiver burden was dichotomized into mild versus moderate-to-severe burden using established CBI cutoff thresholds. Group comparisons, correlation analyses (false discovery rate–corrected), and supervised machine learning models (logistic regression, random forest, AdaBoost, support-vector machine, naïve Bayes, and CatBoost) were performed using 5-fold stratified cross-validation repeated 10 times. Results: Disease-specific burden patterns emerged. Stroke caregivers reported higher time-dependent burden, whereas AD caregivers showed greater emotional burden (p < 0.05). No significant differences were observed in total CBI scores across groups. Lower MoCA scores were moderately associated with higher total and time-dependent burden (r up to −0.49, p < 0.001), particularly when interacting with advanced patient age. Machine learning models showed moderate performance, with AdaBoost achieving the highest accuracy (75.8%) and logistic regression and CatBoost the highest area under the receiver-operating-characteristic curve (AUC = 0.80). The most influential predictors were the age × MoCA interaction, MoCA score alone, and patient–caregiver gender concordance. Cognitive impairment, especially in older patients, and dyadic gender concordance emerged as central risk factors. Conclusions: Multidimensional assessment and early risk stratification, potentially supported by machine learning tools, may improve identification of vulnerable caregivers and guide tailored interventions. Full article

Although new AI systems have been developed based on the integration of information from multiple sources under conditions of uncertainty, classical probabilistic models are unable to provide structured, interpretable, and reliable representations in the presence of contextual and order effects. Specifically, the fundamental principles of Kolmogorov’s assumptions underlying the modeling overlook certain common violations in real-world decision-making processes, such as non-commutativity, contextual dependence among agents, and interaction effects between information sources characterized by experiential heterogeneity. A Positive Operator-Valued Measure (POVM) formalism defined on a Hilbert space of latent states forms the basis of this article’s structured epistemic representation framework to support the reliability and interpretability of the black box in AI. The resulting model generalizes the classical epistemic quadruplet: Probability, Plausibility, Credibility, and Possibility within a single geometric framework in which three essential non-classical effect mechanisms emerge—(i) the non-commutativity of information acquisition, (ii) the contextuality arising from incompatible observational frameworks, and (iii) the interference interactions between information acquisition channels. We propose the concept of a quantum-inspired fusion operator (QI-Happenability), which introduces symmetric and antisymmetric feedback interaction terms based on the estimation of ordered residuals. An analysis of the proposed framework is then performed using real data, validating the model on the Efron et al. diabetes regression dataset (N = 442; ten standardized physiological predictors; continuous disease progression target), available in scikit-learn, which showed a 17.4% reduction in mean absolute error (MAE) compared to traditional models and significant improvements over polynomial machine learning baselines, as well as ensemble machine learning methods, within a rigorous cross-validation protocol. Contextual analysis indicates that 68% of cases violate classical bounds (CHSH inequality, p < 0.001), empirically confirming the non-classical structured representation in multi-source data. The results confirm the proposed approach as a simpler, more interpretable, and more reliable alternative to black-box models: this work demonstrates how the use of structured epistemic representations in reasoning under uncertainty preserves formal interpretability while retaining useful semantic information. By linking quantum cognition and applied AI, this work could help lay the groundwork for a new generation of interpretable and reliable decision-making systems. Full article