Development and validation of a simple prediction model for length of hospital stay in a nationwide Japanese burn registry.

Oceans exhibit complex dynamics influenced by climate change, anthropogenic activities, and natural phenomena. Understanding these dynamics is critical for ensuring the sustainability of marine environments and their optimal utilization. This research aims to study and monitor upwelling phenomena in the South Sea of Java. Upwelling, the exchange of nutrient-rich, cold water from deeper layers to the surface, enhances marine biological productivity; Sea Surface Temperature (SST) serves as a key indicator for its detection. To achieve these objectives, this study employs both ConvLSTM and 3D-CNN. ConvLSTM, a deep learning architecture that integrates convolutional structures within LSTM units, effectively captures spatiotemporal dependencies in sequential data. 3D-CNN, a deep learning model extending traditional 2D convolutional neural networks, processes volumetric data, enabling the extraction of spatial features across three dimensions. Analysis reveals that ConvLSTM outperforms 3D-CNN in modeling upwelling data in the South Sea of Java. This is evidenced by lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). The ConvLSTM method was then used for forecasting, and the results were validated with data obtained from local fishermen regarding their fishing expeditions. Visual analysis confirms that the ConvLSTM method accurately models upwelling data in the South Sea of Java with fishermen's schedules. ConvLSTM and 3D-CNN methods were comparatively evaluated for modeling Sea Surface Temperature (SST) data, considering wind speed, sea surface salinity, and the El Niño-Southern Oscillation (ENSO) phase as influential factors. Based on Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) values, the ConvLSTM method exhibited lower values, indicating superior performance compared to the 3D-CNN approach. Specifically, RMSE and MAE values for ConvLSTM were 0.4161 and 0.3017, respectively, while for 3D-CNN, the corresponding values were 0.6095 and 0.4259. Upwelling data forecasting results were validated against local fishermen's schedules, with data collected in July 2022. Visual inspection confirmed alignment between the forecasted upwelling patterns and the fishermen's activity.

Spatial multi‑criteria site selection for ecotourism suitability mapping: A case study of Mount Abidar, Sanandaj, Kurdistan Province, Iran

Clinical X-ray dark-field radiography has shown to be promising for visualizing different lung pathologies. To keep the radiation dose as low as reasonably achievable (ALARA principle), individualized exposure planning is necessary. However, the current scanning-based implementation of dark-field radiography complicates the use of automatic exposure control. Previously, a BMI-based linear regression model was proposed as a substitute. Here, we aim to improve this proposed model by investigating multiple linear regression for patient-individual exposure planning of dark-field chest radiography. For this retrospective study, 273 posteroanterior thorax images acquired at a prototype system for dark-field chest radiography were analyzed retrospectively regarding the X-ray tube current needed to achieve the target radiation dose. Different multiple linear regression models were tested to find the optimal multiple regression model for predicting the necessary tube current based on a person's weight, height, age, and sex. R2 score, the root mean square error (RMSE), and the mean absolute percentage error (MAPE) were used to evaluate the goodness-of-fit of different regression models. Each model was also compared to a BMI-based model. To predict the target tube current for dark-field chest radiography, multiple linear regression using ordinary least squares performed best (R2 = 0.712, RMSE = 0.234, and MAPE = 0.033). In comparison, simple linear regression using only the body mass index achieved only R2 = 0.627, RMSE = 0.266, and MAPE = 0.037. Multiple linear regression allows better exposure planning in X-ray dark-field chest radiography than simple linear regression.

Comparing allometric models to machine learning models for aboveground biomass estimation in agroforestry systems in Kenya

Robust prediction of parameterized cardiovascular hemodynamics using deep operator networks with time normalization.

Forecasting daily reference evapotranspiration under different hydrological conditions using a hybrid wavelet–Bayesian optimization–Gaussian process regression model

As a result of the emergence of Artificial Intelligence (AI), new applications for measuring active range of motion (AROM) in telerehabilitation (TR) are being developed. The main objectives of the present study were to evaluate the validity of the TRAK® web application for measuring AROM, to assess its reliability as a self-measurement tool for subjects undergoing TR, and to examine the influence of the subjects' technological skills on the self-measurement process. To this end, a cross-sectional observational study was conducted with healthy subjects. Sixty-five volunteer subjects were recruited and divided into two groups based on their technological skills (≤35years and≥55years). Thirteen active joint movements were measured by TRAK® in two sessions, with each session being at least one week apart. Validity was assessed in the first session, during which the AROM data obtained by TRAK® were compared with the data obtained from a subsequent video analysis by Kinovea®. In this first session, the physiotherapist supervised the correct execution of the movement. In the second session, the participants repeated the AROM measurements independently and autonomously, following the instructions given by TRAK®, to analyse the reliability of the tool for TR self-measurement. Regarding validity, TRAK® showed good to excellent correlation (ŕs range=0.739 to 0.987) and root mean square error (RMSE)<4.74°for eleven out of thirteen movements in the younger group. In the older group, TRAK® obtained good to excellent correlation (ŕs range=0.703 to 0.928) and RMSE < 4.95°for ten movements. Concerning reliability, however, TRAK® showed SEM percentages above 10% for multiple movements in both populations in its TR modality. TRAK® proved to be a valid tool for measuring multiple joint movements regardless of the subject's technological abilities, but was unreliable for assessing AROM in TR.

This study proposes a two-level stacking machine learning approach for predicting rainfall erosivity ( R m ) in Taiwan, providing a flexible alternative to traditional empirical methods. Conventional models rely on limited high-resolution rainfall data and are often region-specific, which limits their accuracy elsewhere. In contrast, the proposed ensemble framework captures complex, non-linear interactions among climatic and topographic variables to improve prediction accuracy. In the first level, six base models were combined, and in the second level, each base model was used as a meta-model to form the ensemble structure. Twenty-eight predictor variables, including climatic and topographic factors, were derived from Coupled Model Intercomparison Project Phase 6 (CMIP6) high-resolution global climate data and a digital elevation model (DEM). To ensure robustness, the modeling procedure was repeated five times using different train–test splits, and final performance metrics were calculated as averages across five datasets. Feature selection using Boruta identified rainfall-related variables as the most important contributors. The ensemble approach significantly improved predictive performance, achieving a root mean square error (RMSE) of 5317 . 92 ± 261 . 23 MJ ⋅ mm ⋅ ha − 1 ⋅ hour − 1 ⋅ year − 1 and a Nash–Sutcliffe efficiency (NSE) of 0 . 67 ± 0 . 02 . The analysis revealed an increasing trend in R m , particularly under higher emission scenarios (SSP3-7.0 and SSP5-8.5), with increases projected in the latter half of the century. These findings highlight the importance of targeted climate mitigation and adaptation strategies for soil conservation and watershed management. This study supports Sustainable Development Goals 13 (Climate Action) and 15 (Life on Land) by improving R m prediction to reduce land degradation and enhance climate resilience. • Two-level stacking ensemble ML framework predicts rainfall erosivity ( R m ) in Taiwan. • Combined six base and meta models with 28 climate and DEM predictors. • Random forest (RF) meta-model achieved best accuracy (NSE = 0.67, RMSE = 5317.92 MJ ⋅ mm ⋅ ha −1 ⋅ hour −1 ⋅ year −1 ). • R m shows increasing trends under high-emission scenarios in late 21st century.

Although deep learning reconstruction (DLR) has been shown to improve image quality in MRI, its impact on quantitative physiologic parameters derived from diffusion-weighted imaging (DWI) and dynamic susceptibility contrast (DSC) perfusion in brain tumor imaging remains unclear. To evaluate the impact of DLR on quantitative parameters derived from DWI and DSC in patients with brain tumors. Retrospective. Sixty-two patients (33 male) with post-radiation brain metastasis. 3.0 T; T2, FLAIR, T1WI, DWI, DSC perfusion, and contrast-enhanced T1WI. DWI and DSC images were reconstructed at three DLR levels (high, medium, and low). Agreement between original and DLR images for apparent diffusion coefficient (ADC), cerebral blood volume (CBV), cerebral blood flow (CBF), mean transit time (MTT), and time to peak (TTP) was assessed using the coefficient of variation, repeatability coefficient (RC), and concordance correlation coefficient. For DSC time-series, signal-to-noise ratio, root mean square error (RMSE), and mean absolute error (MAE) were computed within tumor masks. DWI comparisons used mean signal intensity at b = 0 and b = 1000. Paired t-tests compared ADC, relative CBV, and DWI signals. RMSE and MAE were compared using repeated-measures analysis of variance. Significance was set at p < 0.05. ADC (p = 0.955-0.979) and CBV (p = 0.341-0.708), CBF (p = 0.684-0.983), and MTT (p = 0.403-0.971) values showed no significant differences between original and DLR images, while high-level DLR showed significantly higher TTP than original images. RCs demonstrated high reproducibility across DLR levels for ADC (21.78-22.20), CBV (0.88-0.96), CBF (27.98-34.18), MTT (1.26-1.50), and TTP (3.40-3.99). DSC analysis showed the best noise reduction with high-level DLR (lowest RMSE, 254.62 and MAE, 253.18 of DSC) without compromising CBV quantification. DLR effectively reduced noise in DWI and DSC while preserving quantitative accuracy of ADC, CBV, CBF, and MTT. DLR may enable robust physiological MRI when applied in brain tumor imaging. Stage 3.

A data-driven approach for rail temperature estimation from air temperature, solar irradiation, and land surface temperature

Motor Unit Template Estimation Using Integral Shape Averaging.

A sustainable bioremediation approach for treating real contaminated urban river water using natural coagulants from durian and rambutan seeds

Super-Resolution enhanced deep learning for efficient and accurate urban flood simulation at the street scale.

Integration of CFD and machine learning for vehicle cabin thermal management using innovative materials

Quantitative MRI using ultrashort echo time (UTE)-T2* mapping is sensitive to collagen-bound water and tendon microstructure, enabling noninvasive assessment of tendon integrity and laxity. However, the extent to which UTE-T2* measures reflect changes in tendon tension, and their repeatability, remain incompletely understood. We evaluated the sensitivity of mono- and bi-exponential UTE-T2* measures of the Achilles tendon to changes in tendon tension induced by ankle positioning and assessed the test-retest repeatability of these metrics across repeated scan sessions. In this study, healthy adult volunteers underwent UTE-MRI of the Achilles tendon at two scan sessions, spaced 1 week apart, with the ankle positioned in dorsiflexion (higher tendon tension) and plantarflexion (lower tendon tension). Mono-exponential T2* (T2*mono) and bi-exponential parameters (T2*short, T2*long, and short-component fraction ρshort) were quantified. We performed a two-way repeated-measures ANOVA to assess the main effects of ankle position and scan session, and their interaction. Repeatability was evaluated using root mean square error (RMSE), coefficient of variation (CV%), and Bland-Altman analysis. We observed that short-component T2* metrics demonstrated significant sensitivity to tendon tension. A main effect of ankle position was observed for T2*mono (p < 0.001) and T2*short (p = 0.02), with lower values in dorsiflexion compared with plantarflexion. No significant effect of scan session or ankle position × scan session interactions were observed. T2*long and ρshort showed no significant dependence on position or scan session, suggesting that bulk hydration and relative water-compartment contributions remained stable across loading conditions. Dorsiflexion demonstrated lower RMSE and CV% across metrics than plantar flexion, indicating improved repeatability under passive tendon tension. UTE-T2* relaxometry of the Achilles tendon is repeatable and sensitive to changes in tendon tension. Short T2* measures may provide quantitative imaging markers related to tendon mechanical integrity and laxity, while highlighting the importance of standardized tendon tension for longitudinal quantitative tendon MRI.

Spatio-temporal modelling of wind speed using machine learning with a custom Weibull deviance loss for XGBoost

A multi-device data fusion method for 3D rock fracture reconstruction: Development, comparison, and implications for hydro-mechanical simulations

With the increasing incorporation of Renewable Energy Sources (RESs) like solar Photovoltaic (PV) systems, maintaining frequency stability has turned out to be a significant challenge owing to decreased system inertia. Despite numerous developments in Load Frequency Control (LFC), existing solutions largely overlooked the issue of Under Frequency Load Shedding (UFLS) relay failure during rapid frequency decline, which led to widespread blackouts. To address this critical gap, a novel intelligent control framework integrating the Fuzzy Doubleton Parabolic Inference System (FDPIS) and the Proportional Quad-Alpine Integral Derivative (PQAID) controller for UFLS relay failure mitigation and enhanced LFC in Solar-PV systems is proposed. Primarily, the Direct Current (DC) power from the solar module is fed into the DC-DC boost converter and Maximum Power Smoothstep Point Tracking (MPSPT) algorithm. A capacitor bank failure is detected using FDPIS, and voltage stabilization is ensured through a Savitzky-Golay Dynamic Polynomial-Z Voltage Restorer (SGDP-ZVR). To predict electrical load demand accurately, a hybrid Deep Learning (DL) model, Deep Dualplus Softshrink Pan–Long Short Term Inverse Parzen Memory (2DSP-LSTIPM), is employed, delivering a high accuracy of 98.98 % with a Root Mean Squared Error (RMSE) of 0.002. When demand exceeds thresholds, transmission overload is mitigated using an Inductive Snubber Cubic Circuits–STATCOM (IS2C-STATCOM). The frequency deviation is identified via FDPIS, followed by the Rate Of Change Of Frequency (ROCOF) analysis. If a UFLS relay failure is detected, then the PQAID controller is activated to ensure stable operation. The proposed PQAID achieves a peak time of 1.91 ms, significantly outperforming traditional PID, PI, and PD controllers in transient and overshoot metrics. Simulation results on the HEDGW dataset assess the proposed approach’s robustness and low time complexity. The system demonstrates superior relay fault detection (fuzzification/defuzzification times of 452ms/463ms) and faster rule generation (597 ms) compared to conventional fuzzy systems. Overall, the proposed methodology provides a comprehensive, real-time, and scalable solution for enhancing frequency stability, relay fault mitigation, and load management in solar PV-based smart grids. • Integrates FDPIS and PQAID for real-time UFLS relay failure detection and mitigation in solar PV systems. • Proposes novel 2DSP-LSTIPM deep learning model achieving 98.98 % demand prediction accuracy with RMSE of 0.002. • Introduces SGDP-ZVR for voltage stabilization during capacitor bank faults using Savitzky-Golay filtering. • Deploys IS2C-STATCOM for efficient transmission overload control with fast reactive power regulation. • Enables seamless SCADA/EMS integration via OPC-UA protocol for smart grid compatibility and deployment.

• First systematic study quantifying the synthetic-to-real generalization gap in RF fingerprinting localization. • Novel Gaussian Process calibration method significantly improves alignment between simulated and real base station parameters. • Large-scale synthetic pretraining reduces real-world localization error by 50 • Demonstrated that simulation fidelity outweighs dataset size: calibrated synthetic data outperforms larger uncalibrated datasets. Radio frequency (RF) fingerprinting is a promising localization technique for GPS-denied environments, yet it tends to suffer from a fundamental limitation: Poor generalization to previously unmapped areas. Traditional methods such as k -nearest neighbors ( k -NN) perform well where data is available but may fail on unseen streets, limiting real-world deployment. Deep learning (DL) offers potential remedies by learning spatial-RF patterns that generalize, but requires far more training data than what simple real-world measurement campaigns can provide. In this paper, we investigate whether synthetic data can bridge this generalization gap. Using (i) a real-world dataset from Rome and (ii) NVIDIA’s open-source ray-tracing simulator Sionna, we generate synthetic datasets under varying realism and scale conditions. Specifically, we use Dataset A containing real-world measurements with real base stations (BS) and real signals, and create Dataset B using real BS locations but simulated signals, Dataset C with both simulated BS locations and signals, and Dataset B’ which represents an optimized version of Dataset B where BS parameters are calibrated via Gaussian Process to maximize signal correlation with Dataset A. Our evaluation reveals a pronounced sim-to-real gap: Models achieving 25m error on synthetic data degrade to 184m on real data. Nonetheless, pretraining on synthetic data reduces real-world localization error from 323m to 162m; a 50% improvement over real-only training. Notably, simulation fidelity proves more important than scale: A smaller calibrated dataset (53K samples) outperforms a larger uncalibrated one (274K samples). To further evaluate the generalization capabilities of the models, we conduct experiments on an unseen geographical region using a real-world dataset from Oslo. In the zero-shot setting, the models achieve a root mean square error (RMSE) of 132.2m on the entire dataset, and 61.5m on unseen streets after fine-tuning on Oslo data. While challenges remain before meeting more practical localization accuracy, this work provides a systematic study in the field of wireless communication of synthetic-to-real transfer in RF localization and highlights the value of simulation-aware pretraining for generalizing DL models to real-world scenarios.