Search Results (1,579)

Reliable landslide susceptibility mapping (LSM) depends not only on classifier selection but also on the construction of non-landslide samples. Conventional random or buffer-based sampling can retain candidate negatives that are environmentally similar to landslides, increasing label ambiguity and reducing model reliability. This study proposes an IF-KMeans negative sampling framework to refine candidate non-landslide samples for LSM in Wuzhou City, China. Isolation Forest was trained using 395 mapped landslides and then applied to 2000 candidate negative samples to remove samples with high similarity to the landslide feature space; K-Means clustering was subsequently used to stratify the retained candidates and select representative negative samples. The optimized samples were evaluated using six classifiers, including LR, SVM, MLP, RF, XGBoost, and LightGBM, and compared with conventional buffer-based sampling. The IF-KMeans framework consistently improved AUC across the six classifiers, with gains of 0.041–0.081, and the IF-KMeans-RF model achieved the highest AUC of 0.944. Additional diagnostics showed that the IF-removed samples were closer to known landslides in environmental feature space and were located in areas with higher local landslide density, indicating higher potential confusion risk. These findings suggest that positive-sample-guided negative-sample refinement can reduce ambiguity in LSM training data and improve the reliability of susceptibility mapping for geological disaster prevention and risk mitigation. Full article

Background: Retinoblastoma (RB) is a rare pediatric malignancy characterized by small sample sizes and low event rates, where conventional association-driven feature selection may lead to unstable models, overadjustment, and limited generalizability. However, existing survival prediction studies lack a careful treatment of feature selection that accounts for underlying causal structure. Objectives: To develop and validate a causality-guided machine learning model for RB survival prediction by jointly incorporating survival time and survival status as outcome variables. Methods: We analyzed 1015 RB patients from the SEER database (1975–2020). A causality-informed feature selection framework was developed to address the challenges of rare-disease data. Specifically, candidate variables were evaluated through a three-step evidence-integration process: (1) univariate Cox proportional hazards (CPH) analysis for initial statistical screening; (2) causal structure learning using the PC algorithm on the variables retained from Step 1 to construct a directed acyclic graph (DAG) and exclude structurally inappropriate variables (colliders or descendants of the outcome); and (3) LASSO-based feature screening performed independently on the full set of candidate variables. The final features were obtained by taking the intersection of the variables retained from Step 2 and Step 3. Survival models were then trained using the selected features, with model comparison performed as a secondary step. Results: The proposed framework consistently identified four structurally and prognostically robust predictors—laterality, “SEER historic stage A”, “RX Summ”, and sequence number—through this evidence-integration process. Compared with conventional approaches, the causality-informed framework reduced the feature set while improving model stability and interpretability. Notably, compared with LASSO-only selection, which retained a larger set of variables, the causality-informed approach yielded a more parsimonious feature set with improved predictive performance, suggesting reduced overfitting in a low-event setting. Survival models trained on this refined feature set demonstrated reliable predictive performance, with the random survival forest achieving the highest discrimination (C-index = 0.739). Importantly, the selected predictors aligned with clinically plausible pathways in the learned DAG, supporting their causal relevance. Conclusions: This study demonstrates that incorporating causal structure into feature selection provides a more reliable and interpretable foundation for survival modeling in retinoblastoma. Rather than focusing on algorithmic comparison alone, our findings highlight that careful, causality-informed feature selection is critical for improving robustness in rare-disease prediction tasks. This framework may serve as a generalizable methodological template for other rare clinical settings prone to spurious associations. Full article

Data-driven active-suspension models are often judged by aggregate prediction accuracy. For road-adaptive control and diagnostics, however, a model is useful only if its predictions remain reliable when road severity changes. This study asks how far an active-suspension response model trained under one set of ISO 8608 road classes can be transferred to another. A Quanser-based benchmark dataset containing ISO classes A–E is used to evaluate predictability, explainability, and transferability in the same workflow. Sample-level body and tyre displacement models are tested with leave-one-road-class-out validation, while window-level body/tyre RMS and peak indicators are tested through adjacent, graded, and A–D to E transfer scenarios. Transfer loss is quantified with a relative transferability ratio (RTR), defined as the ratio between transfer $R^2$ and in-domain $R^2$. Sample-level tyre displacement transfers more consistently than body displacement, with RTR values of 0.887–0.984 for tyre response and 0.376–0.982 for body response. By contrast, window-level transfer to class E fails for body RMS and peak response and remains weak for tyre RMS, showing that accurate displacement reconstruction does not necessarily translate into transferable engineering indicators. PCA centroid and Wasserstein distances are strongly negatively correlated with RMS RTR, linking transfer degradation to road-feature domain shift. Frequency-band analysis shows that class E produces disproportionate body and tyre response-energy increases in selected low- and mid-frequency bands. SHAP analysis indicates that road RMS dominates tyre–RMS transfer, whereas body–RMS transfer depends mainly on road peak and road RMS. A secondary SOF-controlled dataset is kept separate as a control-effort consistency check. The resulting contribution is an interpretable predictability–explainability–transferability assessment for active-suspension benchmark data, not a new controller, road classifier, or system-identification method. Full article

Background/Objectives: This study aimed to assess technostress and burnout levels among healthcare practitioners, explore the relationships between technostress, perceived organizational support (POS), and burnout, and examine whether POS mediates this relationship using a suppression approach. The rapid proliferation of digital health technologies has increased technology-related job demands in healthcare settings. Technostress has become a major psychological risk factor associated with burnout, but evidence for its direct effects is inconsistent. Although organizational resources, especially perceived organizational support, might moderate this relationship, the underlying mechanism is not well understood, particularly in the context of rapidly digitalizing healthcare systems like Saudi Arabia’s. Methods: This was a cross-sectional, multicenter, descriptive-correlational study with a sample of 150 healthcare professionals conveniently selected from clinical settings in Saudi Arabia. Technostress Creators Scale, Copenhagen Burnout Inventory and Perceived Organizational Support Scale were used for data collection. Statistical analyses included Pearson correlations, multiple regression, bootstrap mediation analyses, and structural equation modeling. Results: Moderate levels of technostress were observed across all dimensions (M = 2.87–3.42). POS was positively related to technostress (r = +0.36, p < 0.001) and negatively related to burnout (r = −0.54, p < 0.001). The bivariate relationship between technostress and burnout was nonsignificant (r = −0.06, p = 0.437). Mediation analysis showed a significant negative indirect effect through POS (β = −0.19, p = 0.002) and a significant positive direct effect (β = +0.14, p = 0.019), indicating an inconsistent mediation (suppression effect). Structural equation modelling demonstrated excellent model fit consistent with a just-identified model. Conclusions: The findings suggest that the observed statistical associations are consistent with a suppression mediation model in which technostress was positively associated with burnout while also being associated with higher perceived organizational support, which in turn was negatively associated with burnout. POS plays a critical protective role but does not fully offset the harmful associations of technostress. Healthcare organizations should implement proactive support strategies, including continuous technical support, structured digital training, and active managerial engagement, to manage technostress and reduce burnout risk. Full article

Purpose: This study examines the preparedness of the Ghanaian construction workforce for Industry 5.0 by assessing digital readiness, identifying skill gaps, and proposing human-centred upskilling strategies. Design/Methodology/Approach: A qualitative approach was adopted, using six focus group discussions with 32 construction professionals from Accra and Kumasi. Participants were selected using purposive and snowball sampling, while data were analysed thematically. Findings: Four major themes emerged: digital readiness, skill gaps, upskilling strategies, and human–machine collaboration. The findings showed that digital readiness was uneven across roles, with design and managerial professionals demonstrating higher exposure to digital tools than site-based workers and supervisors. Six key barriers were identified: limited BIM competence, low digital literacy, poor technological infrastructure, weak organisational support, inadequate structured training, and resistance to technological change. Three upskilling priorities were also identified: role-specific digital training, continuous professional development, and inclusive training models. Originality: The study provides empirical evidence on Industry 5.0 workforce readiness within a developing-country construction context. Practical Implications: The findings support stronger CPD systems, inclusive training programmes, and collaboration among industry, government, academia, and professional bodies. Research Limitations: The study was limited to 32 professionals in Accra and Kumasi; therefore, the findings are context-specific. Full article

Background: Laryngeal squamous cell carcinoma (LSCC) is a highly aggressive malignancy with poor prognosis, particularly in advanced stages. While traditional treatments have improved survival rates, reliable biomarkers for prognosis remain limited. Methods: We analyzed RNA-seq data of LSCC patients from the Cancer Genome Atlas (TCGA) and validated the results using the Gene Expression Omnibus (GEO) dataset (GSE27020), clinical samples, and LSCC cell lines. Differentially expressed immune-related genes (DEIRGs) were identified using the “limma” R package. A prognostic signature was developed by integrating univariate Cox analysis, least absolute shrinkage and selection operator (LASSO) regression, and multivariate Cox analysis. The signature’s predictive performance was validated using Kaplan–Meier survival analysis and receiver operating characteristic (ROC) curves. Results: A three-gene immune-related prognostic signature comprising TNFRSF4, PPARG, and PDGFA was established. In the training cohort, the model stratified patients into high- and low-risk groups with significantly different overall survival (HR = 5.81, 95% CI: 2.56–13.22, p < 0.001), with apparent 1-, 2-, and 3-year AUC values of 0.838, 0.895, and 0.947, respectively. Predictive performance was further evaluated in the TCGA testing cohort, the full TCGA cohort, and the GSE27020 cohort. Functional enrichment analysis revealed that the signature genes are involved in immune regulation and tumor progression. Conclusions: This study identified and validated a novel three-gene immune-related prognostic signature for LSCC, offering a practical tool for individualized prognosis and personalized treatment strategies. The signature provides insights into immune-related mechanisms in LSCC, presenting potential targets for therapeutic intervention. Full article

Accurate prediction of tunnel water inflow in water-rich fault zones is important for groundwater control design and construction risk prevention. In this study, a per-linear-meter tunnel water inflow database containing 425 valid samples was established through orthogonal numerical simulations based on a three-dimensional steady-state seepage model with a grouting ring. The input variables included four hydraulic and grouting parameters and two excavation-position descriptors, namely the excavation-position distance and excavation-position category, thereby reflecting both the water-blocking effect of grouting reinforcement and the spatial variation in water inflow as the excavation face approached the fault zone. Considering that the samples were generated from 25 orthogonal simulation cases at different excavation positions, grouped validation was adopted to reduce information leakage at the simulation-case level. Four baseline machine learning models, including SVM, RF, XGBoost, and CatBoost, were evaluated using ten repeated grouped hold-out validations. CatBoost achieved the best overall baseline generalization performance, with an average test R² of 0.6209 ± 0.0405, MAE of 0.1084 ± 0.0079, and RMSE of 0.1555 ± 0.0085. CatBoost was therefore selected for further hyperparameter optimization. Subsequently, random search, Bayesian optimization, the Osprey Optimization Algorithm, and the Grey Wolf Optimizer were compared under the same search space and computational budget. Hyperparameter optimization was conducted only within the training set using grouped cross-validation, and the independent grouped test set was used only for final evaluation. The results showed that the unoptimized CatBoost model achieved the best overall balance between prediction accuracy, stability, and computational efficiency. Although RS-CatBoost slightly improved MAE and MAPE among the optimized models, none of the optimization strategies consistently outperformed the unoptimized CatBoost baseline, indicating that the choice of hyperparameter optimization algorithm played a secondary role under the current dataset and grouped-validation framework. The proposed framework is intended as a preliminary modeling reference under controlled numerical simulation conditions, and its practical engineering reliability requires further validation using field monitoring data or independent benchmark cases. Full article

Coffee is a major export earner for Uganda, raking in over USD 2 billion in 2025. The global price of coffee is tagged to the perceived quality in the cup which in turn is affected by the chemical composition of the green bean. Breeding for market-preferred Arabica coffee varieties is a major objective of coffee breeding programs. Determination of coffee bean chemical constituents is routinely done through expensive, slow and tedious laboratory procedures, making it unsustainable of resource-limited public sector coffee breeding programs. Here, we demonstrate the use of near-infrared spectroscopy (NIRS) and the machine learning algorithms partial least squares (PLS), random forest (RF) and support vector machine (SVM) for the prediction of caffeine, protein and trigonelline in Arabica coffee. NIRS provides a fast, accurate and reliable method of simultaneously predicting multiple sample constituents. Ripe coffee cherries were picked from 172 farmers’ fields, air dried in the laboratory at room temperature and processed to green beans. NIRS spectra were taken on the milled green bean at 400–2500 nm, with a 0.5 nanometer (nm) step. Reference data for caffeine, protein and trigonelline were collected on the same sample scanned with NIRS. A set of 12 spectral pretreatments were applied prior to making calibrations with the PLS, RF and SVM algorithms and 70% of the data as a training set and 30% as a test set. Caffeine content of reference samples ranged from 1.94–3.0 g/100 g, protein content ranged from 11.16–15.94% while trigonelline ranged from 0.94–1.23 g/100 g. The best calibrations for all algorithms and analytes were obtained using raw (untreated) spectra, which gave the same results as the Savitzky–Golay (SG) pretreatment. For caffeine, the best model (R²p = 0.89, RMSEP = 0.007, RPD = 3.34) was obtained with the SVM algorithm, while for protein, the best model (R²p = 0.98, RMSEP = 0.14, RPD = 6.92) was obtained using the PLS algorithm. Finally, for trigonelline, all three models had very high prediction accuracies (R²p = 0.98–0.99, RMSEP = 0.007–0.009, RPD = 8.53–10.52). Collectively, these results demonstrate the potential of using NIRS for rapid and simultaneous prediction of coffee green bean constituents to aid selection decisions. Full article

Background: Selecting informative training samples is a fundamental yet challenging problem in predictive modeling, particularly in heterogeneous clinical data. Although supervised learning is typically formulated as an optimization problem over model parameters, the composition of the training set substantially influences generalization performance. In this study, we propose a multi-criteria score-based sample selection framework for a machine learning setting. Method: Four sample-level scores were defined to quantify predictive uncertainty, representativeness, non-redundancy, and reliability. These scores were normalized and combined using three integration schemes: additive weighting, reliability-gated weighting, and rank-based aggregation. For each chief complaint category, a baseline model was trained either on the full training set, on score-selected subsets and on random size-matched subsets. Performance was assessed using the area under the receiver operating characteristic curve (AUROC), the area under the precision–recall curve, sensitivity, and specificity, with classification thresholds determined by the Youden index. Results: Across experiments, integrated score-based subset selection outperformed both full-data training and random subsampling in terms of mean AUROC, while often showing lower variability across chief complaints. Conclusions: The results suggest that sample utility in clinical tabular data is intrinsically multi-dimensional and that explicitly modeling this structure can improve predictive discrimination. Full article

With the increasing penetration of behind-the-meter photovoltaic generation and distributed energy storage, distribution system operators usually observe only the net load at the point of common coupling, while the actual user load and energy storage charging/discharging power are difficult to measure directly. To address this problem, this paper proposes a mode-aware constrained inverse optimization method for behind-the-meter distributed energy storage power estimation under fixed time-of-use tariffs. The proposed method uses net load, photovoltaic power, and tariff information as inputs and estimates the hidden user load, storage power, SOC trajectory, and dominant storage arbitrage mode. A mode-aware joint representation model is developed by introducing single-cycle and dual-cycle charge–discharge templates, daily action intensity factors, mode weights, and local correction terms. In addition, power limits, SOC dynamics, SOC bounds, daily energy balance constraints, tariff-response consistency, and mode selection penalty are incorporated into the inverse optimization framework to improve the physical feasibility and interpretability of the estimation results. Case studies are conducted using a 40-day hybrid dataset with a 1 h sampling interval and a 70%/30% training/testing split. The dataset is constructed from park-level user load and photovoltaic data, while the storage power profile is reconstructed according to typical time-of-use arbitrage operation. For the main dual-cycle testing case, the NRMSEs of storage power, user load, and net load are 14.75%, 3.90%, and 3.76%, respectively. The results show that the proposed method can recover the main variation trend of hidden storage power under the studied fixed time-of-use tariff scenario and provides a preliminary basis for park-level storage monitoring and flexible resource perception. Full article

Class-incremental learning allows a model to continuously acquire new classes from sequentially arriving data while preserving its ability to recognize previously learned ones, which is essential for improving adaptability and supporting long-term evolution. However, the class arrival order is inherently random, and highly similar classes may appear consecutively, which intensifies catastrophic forgetting. Although replay-based methods can effectively alleviate this problem, they usually require storing or accessing historical raw samples, which introduces additional data-retention and storage burdens. To address these challenges, this paper proposes ORACIL, an Order-Robust Analytic Class-Incremental Learning framework. First, ORACIL constructs a conflict graph based on class centroids and dynamically partitions newly arriving classes into multiple low-similarity groups, thereby reducing inter-class interference and mitigating forgetting. Second, for each class group, it trains an analytic incremental classification head and performs recursive closed-form updates for the analytic heads using current-stage data and accumulated second-order statistics, without replaying raw historical samples. For group recognition, ORACIL uses feature-derived distance representations rather than raw historical images, making the incremental process raw-sample-free with respect to original image replay. Third, during inference, the group probabilities generated by the group-recognition router are softly fused with the class scores produced by each analytic head, and the class with the highest fused probability is selected as the final prediction. Extensive experiments on CIFAR-100, CUB200, and OmniBenchmark demonstrate the effectiveness of ORACIL. Without replaying historical images, ORACIL achieves final-phase average forgetting rates of 0.16%, 0.77%, and 1.04%, and final-phase accuracies of 95.77%, 93.86%, and 88.12%, respectively. In addition, the MOPD and AOPD results show that ORACIL maintains strong robustness under different class arrival orders. Full article

Side-scan sonar images play an important role in underwater target detection, seabed mapping, and marine environment monitoring. However, the performance of deep learning-based detectors is often limited by the small scale of available sonar datasets, the high cost of data acquisition, and class imbalance among target categories. To address these issues, this paper proposes a diffusion-model-based data augmentation method for side-scan sonar target detection. A FLUX.1 diffusion model is adopted as the base generative framework and is fine-tuned using low-rank adaptation (LoRA) to adapt the pretrained model to the side-scan sonar image domain under limited training data conditions. The generated samples are further filtered and added only to the training set, while the validation and test sets are kept unchanged and contain only real sonar images. To ensure a fair evaluation of the augmentation strategy, all detection experiments are conducted using a fixed YOLOv8n (You Only Look Once version 8 nano) detector under the same training hyperparameters and three random seeds. Compared with training on the original dataset, the proposed FLUX+LoRA augmentation improves mean average precision (mAP)@0.5 from 0.7400 ± 0.0132 to 0.8582 ± 0.0328 and mAP@0.5:0.95 from 0.3994 ± 0.0187 to 0.5115 ± 0.0164. It also outperforms conventional augmentation methods under the same real-only validation/test protocol. In addition, Fréchet Inception Distance (FID)/Kernel Inception Distance (KID)-based image quality evaluation, generated-sample amount ablation, screening-strategy ablation, LoRA-rank sensitivity analysis, and a controlled 600-sample diffusion-backbone comparison are conducted. The results show that the 600-sample manually annotated FLUX+LoRA subset selected from generated samples achieves better image quality and detection performance than FLUX-base and SD1.5+LoRA under the same annotation budget. These findings demonstrate that FLUX+LoRA-generated sonar images can provide useful structural diversity for detector training and improve target detection performance under limited-data conditions. Full article

Machine learning (ML) is increasingly applied to geochemical baseline estimation and anomaly detection in soils and sediments, yet the methodological conditions under which machine learning outperforms traditional approaches—and which preprocessing and validation decisions most consequentially determine that advantage—remain incompletely characterized across environmental and mineral exploration domains. A structured systematic scoping review of 146 records from the Web of Science Core Collection applied sequential filtering to yield 78 thematically eligible studies, from which 20 were prioritized through a composite index integrating age-adjusted citation impact, platform usage, and semantic relevance. Four cross-cutting findings emerge. First, performance gains in environmental applications were driven primarily by spatial model structure rather than algorithm selection: incorporating a spatial covariate derived from geographically weighted regression raised test-set explained variance from $R^2=0.80$ to $R^2=0.96$ for cadmium mobility prediction in a geochemically heterogeneous karst setting, a gain the source study supported with a held-out test set and a Monte Carlo analysis of sensitivity to data size. Second, isometric or centered log-ratio preprocessing was applied in the majority of mineral exploration studies (three of five classical and hybrid studies and four of five deep-learning studies) but in none of the seven environmental studies, representing a systematic methodological gap with direct consequences for covariate importance estimates under compositional closure. Third, Shapley additive explanations and accumulated local effects functioned as instruments of operational value, enabling element-specific anomaly threshold derivation, training sample diagnosis, and grid-cell anomaly type classification; this evidence demonstrates that the accuracy–interpretability trade-off commonly assumed in the machine learning literature is not fundamental in geochemical applications but contingent on algorithm selection. Fourth, 90% of the 20 synthesized studies (18 of 20 by study-area location—13 in China and five in Iran) were evaluated under within-domain validation designs, and the consistently high performance metrics reported should be interpreted as interpolation estimates rather than evidence of transferable predictive capability. Geographic diversification of training datasets and spatially explicit cross-regional validation are identified as structural prerequisites for regulatory-grade applicability. Full article

Accurate single-step prediction of stack NO_x concentration is essential for emission monitoring and ammonia-injection control in cement kiln SNCR–SCR hybrid denitrification systems. However, this task is challenging because industrial kiln data are affected by nonstationary emission fluctuations, nonlinear multivariable coupling, process-dependent time delays, and online deployment constraints. To address these process-specific challenges, this study develops a leakage-free dynamic soft-sensing framework for stack NO_x concentration prediction. In the proposed framework, variational mode decomposition (VMD) is used to characterize the multi-scale nonstationarity of the stack NO_x sequence under a sliding-window protocol. Trend-guided maximal information coefficient (MIC) analysis is then applied for nonlinear feature selection and delay compensation using only the training data, and the identified feature subset and delay parameters are fixed for validation and testing. A TCN-SE-LSTM model is constructed to extract temporal dependencies, recalibrate informative feature channels, and capture long-lag dynamic behavior. In addition, the Dual Adaptive Constrained Ivy Algorithm (DAC-IVY) is used only for offline hyperparameter optimization, so that the online stage requires only the trained prediction model. Experiments using 21,600 raw samples collected from an actual cement kiln Distributed Control System (DCS) show that the proposed framework achieves an RMSE of 0.2084 mg/Nm³ and an $R^2$ of 0.9844 on the test set, outperforming conventional baseline models. These results indicate that the proposed framework can provide an effective soft-sensing basis for subsequent denitrification control and operational optimization. Full article

Accurate and transferable estimation of forest carbon is essential for operational forest management and national greenhouse gas reporting, yet it remains challenging because of variation in stand structure and site conditions. Unmanned aerial vehicle-based light detection and ranging (UAV-LiDAR) provides detailed structural information for modelling tree-level carbon, but model transferability across sites is often limited. In this study, we compared three modelling approaches—a linear mixed-effects model (LMM), a generalised additive model (GAM), and Random Forest (RF)—within a unified framework of multi-site, locally post hoc calibrated, and fully local model-fitting strategies. Using data from 20 radiata pine (Pinus radiata D. Don) plantation stands across New Zealand (35,201 trees), a leave-one-site-out (LOSO) framework was used to assess multi-site model transferability and support post hoc calibration, while local models were evaluated using repeated within-site train/test splits. We also evaluated how prediction accuracy changed with increasing local sample size and compared random tree selection with plot-based sampling. Multi-site models showed poor generalisation, with mean relative RMSE ranging from 35.9% to 56.9% and substantial site-level bias. Applying post hoc calibration to the multi-site model using a 50-tree sample reduced prediction error by 30 to 60% (mean relative RMSE 22.8–25.0%) and substantially reduced bias across sites. The fitting of fully local models with the same sample size yielded only modest further improvements (mean relative RMSE 21.9–23.1%). Gains in accuracy were minimal with increasing sample sizes above 50 trees for post hoc calibration and 175 trees for the fully local models, and differences in accuracy between sampling strategies were small. These results show that post hoc calibration of multi-site UAV-LiDAR models with a small local sample provides a practical and efficient approach for tree-level carbon estimation in plantation forests. Full article