Search Results (6,621)

In the context of urban regeneration, identifying the nonlinear and interactive effects of the built environment on commercial vitality is essential for targeted spatial improvement. Using Xi’an’s central urban area as a case study, this study integrated multi-source data, including POI, AOI, street-view imagery, and mobile phone signaling data, to delineate commercial spaces via kernel density analysis. With actual service population density as the vitality indicator, a built-environment framework was constructed using 14 indicators across four dimensions: transport accessibility, functional diversity, street quality, and environmental capacity. Random forest regression and SHAP-based interpretable machine learning were employed to examine factor importance, nonlinear thresholds, and interactions. Results show that environmental capacity and transport accessibility are the dominant dimensions, with building density, road network density, and employment density contributing most. Built-environment variables generally exhibit nonlinear threshold effects; key thresholds include road network density > 8 km/km², building density > 40%, functional mix > 4.5, and sky view factor around 40%. Interactions involving building density are most pronounced, and its positive effect is significantly amplified under higher accessibility or employment density. These findings suggest prioritizing road network optimization and building coverage, while balancing functional mix and spatial scale in commercial space regeneration. Full article

The growing worldwide prevalence of Diabetes Mellitus highlights the urgent need for effective early detection methods to enable prompt intervention. This study develops a machine learning-based decision-support prototype for predicting diabetes risk using health metrics from the DiaHealth dataset, a recently published Bangladeshi open-source dataset for Type 2 diabetes prediction. Five supervised learning algorithms were evaluated: Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Decision Tree (DT), and Random Forest (RF). Models were assessed across three stages: before feature scaling, after standardisation, and following hyperparameter optimisation via GridSearchCV, using accuracy, precision, recall, and F1-score as evaluation metrics. LR and SVM showed marked improvements after standardisation, consistent with their sensitivity to feature magnitude, whilst tree-based approaches such as DT and RF remained largely unchanged. KNN displayed minimal sensitivity to scaling, which is discussed in relation to the feature distributions of the dataset. Following hyperparameter tuning, RF achieved the highest accuracy of 95%, outperforming all other models. RF predictions were interpreted using Local Interpretable Model-agnostic Explanations (LIME) to promote transparency in model decision-making. The best-performing model was subsequently deployed as an interactive web-based prototype application using Streamlit, providing real-time prediction outputs. These findings demonstrate how preprocessing choices and hyperparameter tuning can differentially affect algorithm performance and illustrate the potential of combining explainable AI with practical deployment for diabetes risk assessment in a research context. Full article

The large-scale integration of wind and solar power introduces significant volatility into electricity markets, posing challenges for accurate day-ahead price forecasting for generation companies. This paper proposes a hybrid forecasting model, CEEMD-SE-IBA-LSTM, based on hyperparameter adaptation to improve prediction accuracy. First, a similar-day selection method integrating Random Forest and an Improved Grey Ideal Value approximation identifies the most relevant historical days. Second, Complete Ensemble Empirical Mode Decomposition with Sample Entropy (CEEMD-SE) decomposes and reconstructs the price series into stable components. Third, an Improved Bat Algorithm (IBA), incorporating differential evolution and adaptive weighting, is developed to optimize two key LSTM hyperparameters: the number of hidden layer neurons, which is treated as a model architecture hyperparameter, and the learning rate, which is treated as a training hyperparameter. The number of LSTM layers and the number of training epochs are kept fixed as model settings to ensure reproducibility. Using data from the US PJM market, the proposed model is validated against six benchmarks. The results show that CEEMD-SE-IBA-LSTM achieves superior performance, with a Mean Absolute Percentage Error (MAPE) of 3.73%, a Root Mean Square Error (RMSE) of 3.57 $/MWh, and a Mean Absolute Error (MAE) of 1.95 $/MWh. The method provides accurate price trends, offering effective decision support for new energy enterprises in price bidding to enhance revenue. Full article

Radiomics has the potential to improve risk stratification in head and neck squamous cell carcinoma (HNSCC), but clinical adoption is limited by inconsistent performance across institutions. A key source of variability is how radiomic features are generated, preprocessed, and selected prior to model development. This multicenter study evaluated how radiomics parameterization and feature selection strategies affect external model performance, feature stability, and time-to-event risk stratification. We studied pre-treatment CT scans from 752 patients with primary HNSCC from three hospitals. For each scan, 1648 radiomic features were computed using 20 different preparation methods that varied in scaling, outlier removal, and gray-level bin width. We compared five feature selection methods: Graph-FS with connected components, Boruta, Lasso, RFE-RF, and mRMR. The classification models used were Random Forest, XGBoost, CatBoost, and Logistic Regression. We measured performance using external ROC-AUC, bootstrap confidence intervals, Brier score, and RobustScore. Stability of feature selection was assessed using the Kuncheva and Jaccard indices. Cox proportional hazards models confirmed time-to-event results, and consensus SHAP analysis helped explain the models. Radiomics parameterization influenced model performance, and no single configuration was optimal across all analyses. Radiomics-only models outperformed clinical-only models, while clinical–radiomics models achieved the highest overall performance. mRMR and Lasso produced the highest average external AUCs, while Graph-FS showed the greatest stability. The best classification model achieved an external AUC of 0.817. In Cox validation, the best clinical–radiomics configuration achieved an external C-index of 0.662 and separated high- and low-risk patients in the external cohort. Full article

The proposed methodology combines analytical control laws with adaptive mechanisms and machine-learning-assisted modules based on regression trees, random forests, and extreme gradient boosting (XGBoost). Machine learning models are employed to approximate unknown nonlinear dynamics, compensate disturbances, and adjust controller parameters, while the overall control structure is constrained by Lyapunov stability conditions. This ensures that the inclusion of data-driven components does not violate the fundamental requirement of system stability. The effectiveness of the proposed approach is evaluated through simulation experiments across three operating modes with varying degrees of nonlinearity and dynamic complexity. The results show that hybrid models incorporating ensemble machine learning methods improved performance compared with the analytical and adaptive baselines examined. XGBoost-based control achieves the lowest error values and the highest level of Lyapunov stability compliance (up to 99.3%). The main contribution of this study lies in the development of a unified synthesis framework in which machine learning is not used as a standalone control strategy but as a machine-learning-assisted support mechanism integrated into a theoretically grounded control architecture. The proposed approach provides a balance between adaptability, accuracy, and rigorous stability guarantees, suggesting potential applicability to simulation-based and offline-assisted control design tasks, while real-time embedded implementation requires additional computational optimization and validation. Full article

This study evaluated the effects of maternal overfeeding during mid-to-late gestation on maternal productivity, metabolic status, reproductive recovery, and calf performance in Hanwoo cattle using conventional statistics and machine learning (ML) approaches. A total of 243 pregnant cows were assigned to either a control group or an overfeeding group from gestation day 90 to parturition. The overfeeding treatment increased nutrient supply to approximately 140–145% of the control level. Maternal body weight (BW), body condition score (BCS), serum metabolites, and reproductive traits were evaluated throughout gestation and postpartum, while calf growth, morphometrics, and metabolic traits were assessed at birth and weaning. Calves were further classified into growth- or meat-quality-oriented genotypes using SNP-based profiling. Overfeeding increased maternal BW gain and BCS during gestation and reduced circulating non-esterified fatty acid concentrations, indicating improved maternal energy status. However, overfed cows showed a longer interval to postpartum estrus return. Calf birth weight was not significantly affected by maternal overfeeding, whereas calf growth and morphometric traits at weaning were more strongly influenced by parity, sex, and genotype. Machine learning models identified gestational BW, metabolic indicators, calf feed intake, and genotype as major predictors of maternal and calf outcomes, with random forest and XGBoost showing superior predictive performance compared with linear models. These findings suggest that parity- and genotype-informed nutritional management combined with ML-based prediction may support precision feeding strategies in beef cattle production systems. Full article

The development of Software-Defined Networks (SDNs) introduces new challenges in network security, particularly in detecting Distributed Denial of Service (DDoS) attacks and network anomalies. Due to the centralized architecture of SDN, traditional detection methods are often insufficient in dynamic environments. Therefore, machine learning techniques are increasingly applied to improve detection effectiveness. This paper analyzes the impact of network topology on the performance of machine learning-based detection methods in SDN environments. A controlled experimental setup based on the RYU controller and OpenFlow 1.3 was implemented using Mininet. Two network topologies (linear and hierarchical) were evaluated under multiple attack scenarios, including TCP SYN flood and TCP/UDP port scanning. Two supervised learning models, Random Forest (RF) and K-Nearest Neighbors (KNN), were implemented and compared using standard evaluation metrics: accuracy, precision, recall, F1-score, and detection time. The results show that Random Forest significantly outperforms KNN, achieving up to 100% accuracy and detection times as low as 4.24 s, while KNN exhibits lower stability and reduced recall in anomaly detection scenarios. The study demonstrates that network topology has a measurable impact on both detection performance and latency. The observed effects varied across attack scenarios and machine learning models. Hierarchical topology generally improved detection sensitivity in DDoS scenarios, while linear topology often enabled lower detection latency during selected anomaly detection experiments. The results indicate that both machine learning model selection and network topology should be jointly considered when designing intrusion detection systems for SDN environments. These findings contribute to improving the effectiveness and responsiveness of security mechanisms in modern programmable networks. Full article

Accurate obesity risk classification is often hindered by the asymmetric and heteroscedastic nature of health data, where traditional mean-based machine learning models fail to capture critical distribution tails. This study addresses this gap by proposing a robust Expectile Random Forest (ERF) model, a novel ensemble architecture that integrates an expectile learning framework via the Asymmetric Least Squares (ALS) loss function for seven-level (multilevel) classification. Utilizing a dataset of 2111 empirical records, the sensitivity analysis identifies $\tau =0.7$ as the optimal configuration, achieving an overall Accuracy of 94.6 ± 0.7% and a Macro F1-Score of 94.5 ± 0.7%. This performance represents a significant quantitative improvement over state-of-the-art benchmarks, outperforming XGBoost by 1.8% and standard Random Forest by 3.9%. Feature importance analysis identifies body weight, age, and sedentary factors as primary predictors, while the ERF model demonstrates exceptional ordinal consistency and robustness against clinical outliers. These findings provide a superior methodological framework for developing precise medical decision support systems, shifting the paradigm from central-tendency predictions to tail-sensitive health risk mapping. Full article

The mean stand diameter at breast height (DBH) is a key indicator of stand structure and productivity and is widely used in forest resource inventory and management planning. When using regional inventory data, nonlinear interactions between plot-level conditions and predictor variables can undermine the stability of traditional empirical equations across varying site qualities and stand densities. To improve the accuracy and robustness of inventory-scale predictions of mean stand DBH, this study utilized data from 854 forest plots and employed stand age, site class index (SCI), and stand density index (SDI) as independent variables. The predictive performance of traditional growth equations, machine learning models (Random Forest, XGBoost, LightGBM, and support vector machine), and deep learning models (MLP and CNN, ResNet, RNN) was systematically compared, and ensemble learning strategies were further applied to optimize model performance. The results indicated that the Weibull model based solely on stand age achieved the best fit (R² = 0.669). Incorporating SCI and SDI greatly improved model explanatory capability with R² rising to 0.838. XGBoost and CNN further improved predictive performance (R² = 0.852 and 0.861, respectively), while the ensemble model exhibited the highest goodness-of-fit (R² = 0.893), outperforming all individual models. Compared with linear regression, machine learning models demonstrated superior predictive capability. A feature importance analysis indicated that stand age, site quality and stand density together drive mean stand DBH prediction, among which stand age and stand structural characteristics are the dominant influencing factors, whereas SCI and SDI have comparatively weaker effects. Overall, the ensemble model substantially enhanced the prediction accuracy of mean DBH in Pinus yunnanensis stands, thereby providing for precision forest management and ecological function assessment. Full article

Introduction: River fragmentation caused by hydropower infrastructure remains a primary threat to aquatic biodiversity, creating a critical need for fish passage solutions that can adapt to high environmental variability. Although adaptive management (AM) has the potential to significantly improve longitudinal connectivity and ecological resilience, its application in real-world fishway operations is currently limited. Objective: This study aims to present and validate a flexible AM framework designed to optimize fish passage by integrating low-cost monitoring systems with automated data processing and predictive modeling. Methodology: The proposed system combines a sensor network for real-time water-level and environmental monitoring with biological performance data obtained through Passive Integrated Transponder (PIT) technology. These data were processed locally using edge computing. Over a two-year period, weekly aggregated data were used to develop Random Forest models to identify the primary drivers of fish movement. Results: The final model successfully identified five key drivers: luminosity, water temperature, and three nested hydraulic parameters at the fishway’s upstream section. Validation at a vertical-slot fishway in Vadocondes (Duero River, Spain) showed that retrospective optimization—specifically adjusting sluice-gate regulation—could increase downstream water levels and reduce drops at the first cross wall. This adjustment demonstrated a substantial increase in predicted fish passage without requiring changes to the hydropower plant’s core operation. Conclusions: The framework is highly flexible and transferable to other regulated river systems. However, its success is contingent upon the definition of clear ecological objectives and the seamless integration of monitoring results into the day-to-day operation of river infrastructure. Full article

Estimation of rice crop yield at the village level is essential because village is the Insurance Unit (IU) for rice crop in many regions in India, and timely and accurate yield information at this scale supports timely and transparent claim settlements for farmers and supports local agricultural planning. To achieve this, a multi-source satellite data-based machine learning approach was used to estimate rice yield at the village level using optical and SAR data, climatic data and land surface model-derived parameters in Kakinada of Andhra Pradesh, India. The predictor dataset included seasonal cumulative rainfall, seasonal Normalized Difference Vegetation Index (NDVI)-Max, seasonal NDVI-Mean, seasonal Land Surface Water Index (LSWI)-Max, seasonal LSWI-Mean, season total Fraction of Absorbed Photosynthetically Active Radiation (fAPAR) and season total Root Zone Soil Moisture (RZSM), and season total backscatter of the Sentinel-1 VH polarization were used to represent crop greenness, moisture status, photosynthetic activity, soil water availability, canopy structure, and seasonal water supply. For model development and validation, village-level rice yield data from 2017 to 2023 was used, which was collected through Crop Cutting Experiment (CCE) at the maturity stage of Kharif season. In this study, four machine learning models such as Random Forest (RF), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and Gradient Boosting (GB) were evaluated. The multi-source satellite data and yield data for the period 2017–2021 were used to train the models, which were independently tested on 2022 data and then applied to predict the rice yield in 2023. Leave-One-Year-Out (LOYO) cross-validation was also conducted on the 2017–2022 data to assess temporal robustness and generalization capability across years. Among the evaluated models, Random Forest exhibited the best overall performance. For the independent test year 2022, RF achieved an R² of 0.465, RMSE of 415.34 kg ha⁻¹, MAE of 322.22 kg ha⁻¹, and MAPE of 10.36%. For the prediction year 2023, RF achieved improved accuracy with an R² of 0.838, RMSE of 325.75 kg ha⁻¹, MAE of 262.21 kg ha⁻¹, and MAPE of 7.68%. Further, LOYO cross-validation also showed the robustness of RF, achieving the highest mean R² of 0.702 and mean RMSE of 384.73 kg ha⁻¹. The results illustrate that multi-source satellite data combined with machine learning can be a reliable and operationally useful tool in predicting village-level rice yield, which can be used for crop insurance claim settlement. Full article

The Robot Operating System (ROS) is widely used in modern robotics, but its open architecture makes it vulnerable to numerous cyber threats. Although machine learning (ML)-based intrusion detection systems (IDSs) demonstrate strong classification performance on ROS-specific datasets, reliance on topology-dependent identifiers such as source and destination IP addresses, port numbers, and Flow IDs remains a critical limitation in current research. This reliance may encourage algorithms to exploit scenario-specific endpoint signatures instead of relying primarily on transferable behavioral patterns. Consequently, classification scores may be artificially inflated due to data leakage. This study addresses this issue by quantitatively measuring the impact of data leakage and introducing a topology-independent, explainable ROS framework that provides a more realistic, leakage-aware, and topology-independent evaluation framework. The evaluation involved testing the LightGBM, XGBoost, and CatBoost algorithms on ROSIDS23. Additionally, Random Forest and Gradient Boosting were included to verify the presence of data leakage. In our ablation study, models that included topology features achieved near-perfect Macro-F1 values of 0.999 to 1.000. In contrast, removing topology-dependent features reduced the Macro-F1 score to about 0.66. This finding shows that topology descriptors, rather than just transferable attack behaviors, can significantly influence the near-perfect scores seen with topology-preserving protocols. Even without topology data, ML models effectively captured temporal behavioral patterns and detected DoS attacks with nearly perfect performance, reaching F1 scores of 0.99 or higher. However, semantic attacks like Unauthorized Subscribe remained tough to classify, with F1 scores of 0.43 or lower. Additionally, SHapley Additive exPlanations (SHAP) analysis improves the interpretability of IDSs by identifying the main behavioral features that drive model decisions and suggesting feature-level directions for rule-based defense configurations in ROS environments. Full article

Background: Inter-individual variability in lipid response to dietary interventions complicates cardiometabolic prevention in overweight and obese women. Although several dietary strategies improve lipid profiles on average, predictors of comprehensive, multi-marker lipid improvement remain unclear. Objective: To identify baseline clinical predictors of comprehensive lipid improvement across seven dietary interventions and to evaluate the performance of three machine learning (ML) classifiers in predicting a composite Global Score. Methods: This secondary analysis pooled individual-level data from 284 overweight or obese women enrolled in three randomized controlled trials (RCTs). Participants were assigned to continuous energy restriction (CER), intermittent energy restriction (IER), intermittent energy and carbohydrate restriction (IECR), IECR with added protein and fat (IECR+PF), high-carbohydrate, high-monounsaturated-fat, or daily energy-restriction diets. Eleven baseline clinical features served as predictors. Four binary lipid improvement scores (TC/HDL, LDL/HDL, non-HDL cholesterol, TG/HDL) were calculated from baseline to week 12, and a composite Global Score was defined as TRUE only when all four improved. Three ML classifiers (J48, Logistic Model Tree [LMT], Random Forest) were evaluated using stratified 10-fold cross-validation. Results: Overall, 30–35% achieved improvement in the Global Score. Improvement rates varied across diets, with High Mono and High Carb showing the highest rates (48.4%). LMT performed best (AUC = 0.66; accuracy = 70%). Baseline TG, BMI, age, total cholesterol, and weight were the strongest predictors. Conclusions: Comprehensive lipid improvement varies across dietary strategies and is influenced by baseline triglycerides, adiposity, age, and diet type. ML-based stratification may support personalized dietary prescriptions. Full article

Ground improvement for foundations supported on soft soils is traditionally problematic because of low bearing capacity and a large magnitude of settlement. One sustainable method for mitigating these problems is the use of stone columns (SCs), particularly geosynthetic-encased stone columns (GESCs), to improve load transfer, confinement, and consolidation. This review critically synthesizes recent advances in the analysis and design of SC systems using experimental investigations, numerical simulations, and machine learning (ML)-based methodologies. The article indicates that GESCs, when integrated with modern data-driven techniques, especially hybrid metaheuristic ML models, represent a reliable and sustainable solution for soft soil stabilization. Traditional analytical and empirical methods remain useful; however, they are often inadequate for very soft soils (Undrained shear strength (c_u) < 15 kPa), where excessive bulging and large deformations dominate system behavior. Consequently, intelligent hybrid modeling approaches are emerging as the next generation of optimized, data-driven design tools in geotechnical engineering. Different failure mechanisms of SCs, including bulging, punching shear, and general shear failure, are critically discussed along with the governing design parameters. Previous studies consistently indicate that spacing ratios within the range of s/D = 2–3 can improve the bearing capacity ratio (BCR) by approximately 50–100%. Numerical and experimental studies further demonstrate that SC systems can transfer nearly 60–80% of the applied load through stress concentration and soil arching mechanisms. Furthermore, the application of geosynthetic encasement enhances the performance of SCs in very soft soils by increasing confinement, reducing lateral deformation, and enhancing bearing capacity by nearly 3–6 times compared with ordinary SCs. The review also evaluates the growing role of artificial intelligence techniques in forecasting settlement and bearing capacity behavior. ML techniques such as artificial neural networks (ANN), support vector regression (SVR), random forest (RF), XGBoost, and hybrid metaheuristic–ML models have shown high predictive capability, often achieving prediction errors below 5%. Despite these advancements, many existing ML studies still suffer from limited datasets, a lack of generalization, and insufficient incorporation of physical mechanisms. Full article

Dengue virus infection remains a significant public health challenge in endemic regions, with growing evidence of autochthonous transmission in Europe. Assessing serological exposure in high-risk populations such as military personnel deployed to endemic areas is essential to quantify exposure risk and support operational decision-making, particularly regarding pre-deployment counselling and risks associated with secondary infection. We conducted a cross-sectional study involving 1355 members of the Italian Armed Forces, measuring anti-dengue IgG antibodies by ELISA and collecting data on deployment history and exposure risk. Overall, IgG seropositivity was 8.12%, with significantly higher prevalence among individuals reporting travel or deployment to endemic regions (24.71%) compared with non-exposed personnel (4.27%). Seropositivity increased with age and correlated with a CDC-derived cumulative dengue risk score (Spearman’s ρ = 0.299, p < 0.0001). A multivariable logistic regression model including age and exposure risk achieved an AUC of 0.75, while machine-learning models provided complementary predictive assessment, with random forest reaching an AUC of 0.79. These findings indicate substantial anti-dengue IgG seropositivity compatible with previous dengue exposure among Italian military personnel, particularly those deployed to endemic settings. The study highlights the need for targeted surveillance and risk-based preventive strategies, and supports the use of exposure-based models to improve epidemiological assessment and inform vaccination strategies in mobile populations. Full article