Purpose Relative density (RD) is a key quality indicator in laser-based powder bed fusion (L-PBF), linked to microstructure, mechanical properties and performance. This study aims to improve the prediction of RD by integrating a wider set of continuous and categorical inputs, capturing multifactorial interactions beyond the process parameters. Design/methodology/approach A data set of 1,579 samples was compiled from 85 peer-reviewed studies, covering multiple alloys, atmospheres, geometries and measurement methods. Exploratory data analysis combined mutual information and correlation metrics to assess feature relevance. K-means clustering segmented the data into homogeneous groups. Within each cluster, ensemble learning models were optimized via grid search and metaheuristics, with performance validated against literature and experimental data. Findings The cluster-driven framework achieved high predictive accuracy (R2= 0.94) across alloys and process ranges. Clustering improved generalization, especially in low-density regimes. Feature relevance varied by cluster: powder D50, geometric factor and laser power consistently ranked highest. Gradient boosting performed best in some clusters, while weighted-sum and voting ensembles provided the most balanced accuracy. SHAP analysis revealed complex, nonlinear interactions among geometric and process parameters. Originality/value This work introduces several novel contributions to the prediction of RD in L-PBF: the expansion of the input feature space to include underused variables such as material, shielding atmosphere, geometric descriptors and a newly defined shape factor; the use of a cluster-specific modeling strategy (“cluster-then-model”) that tailors regressors to data subgroups based on process-response similarity; and the integration of dual-ensemble optimization with explainability methods, resulting in a robust, transferable and interpretable framework for process performance prediction in metal additive manufacturing.

Invasive freshwater turtles are major drivers of biodiversity loss, underscoring the importance of early detection and management. However, it is challenging for experts to manually monitor a broad geographic area, necessitating support tools. Deep learning-based object detection models have displayed high performance in automating wildlife monitoring tasks. Furthermore, hyperparameter optimization, including optimizer selection and hyperparameter tuning, might further enhance performance by optimizing training settings to the dataset. In this study, an optimized model was developed to apply hyperparameter optimization to detect and classify six invasive turtle species in Korea from images. The optimized model was compared to a default model trained using the default optimizer and hyperparameters. The optimized model outperformed the default model, as indicated by the evaluations of mean average precision using a fixed intersection over union threshold of 0.5 (0.973 vs. 0.959) and a range of thresholds ranging from 0.5 to 0.95 (0.841 vs. 0.815). The classification accuracy of the optimized model reached 92.7%, exceeding that of the default model (89.9%). These findings highlight the utility of hyperparameter optimization and suggest that the proposed approach can support the early detection of invasive turtles, thereby enhancing to invasive species management.

Wire explosion (WE) inherently generates particle ensembles spanning the nano- to microscale, posing challenges for conventional characterization methods in terms of capturing the full particle population. To address this issue, spectrophotometric analysis combined with algorithmic spectrum reconstruction based on Mie theory and constrained distribution models were employed to characterize copper WE products formed in aqueous surroundings within the 4–12 kV discharge voltage range. Three independent fitting strategies, specifically a semimanual fitting, an evolutionary algorithm, and a grid search, were applied to retrieve the size distributions and relative shares of copper and copper oxide particles as a function of discharge voltage. Based on experimental and theoretical findings, lognormal and normal distributions across the 10–300 nm diameter range were assumed as constraints for oxide and metallic fractions, respectively. The reconstructed metallic copper population exhibited mean diameters ranging from 123 to 181 nm, while oxidized fractions followed lognormal distributions centred near 10 nm mode diameters. Voltage-dependent trends revealed an optimal discharge regime between 6 kV and 8 kV, where the exploded fraction reached approximately 63% and the metallic mass share exceeded 80%. These results confirmed that spectrophotometry represents an essential tool for the quantitative characterization of such complex, wide-range systems.

Radar image extrapolation serves as a fundamental methodology in meteorological forecasting, facilitating precise short-term weather prediction through spatiotemporal sequence analysis. Conventional approaches remain constrained by progressive image degradation and artifacts, substantially limiting operational forecasting reliability. This research introduces E-HEOA—an enhanced deep learning architecture with integrated hyperparameter optimization. Our framework incorporates two fundamental innovations: (a) a hybrid metaheuristic optimizer merging a Gaussian-mutated ESOA and Cauchy-mutated HEOA for autonomous learning rate and dropout optimization and (b) embedded ConvLSTM2D modules for enhanced spatiotemporal feature preservation. Experimental validation on 170,000 radar echo samples demonstrates superior performance, demonstrating considerable enhancement in almost all aspects relative to the baseline model while establishing new state-of-the-art benchmarks in prediction fidelity, convergence efficiency, and structural similarity metrics.

Patients with locally advanced rectal cancer (LARC) who undergo neoadjuvant chemoradiotherapy (NCRT) and subsequently experience early recurrence (ER) within two years post-surgery tend to have unfavorable prognoses. Therefore, the accurate prediction of ER in LARC is of paramount importance. This study aimed to develop and validate an explainable artificial intelligence (AI) model, based on the systemic inflammation-nutritional tumor biomarker index (SINTI) derived from routine blood biomarkers, to predict ER in patients with LARC. We conducted a multicenter retrospective analysis involving two distinct patient cohorts: Cohort A (n = 715; from February 2011 to September 2017) and Cohort B (n = 224; spanning June 2020 to June 2023). Feature selection was executed utilizing the least absolute shrinkage and selection operator (LASSO) regularization to construct SINTI, effectively addressing multicollinearity. Predictive modeling incorporated ten different machine learning architectures, with hyperparameter optimization achieved through a randomized search complemented by nested tenfold cross-validation. Model performance was thoroughly evaluated using multiple metrics, including the area under the receiver operating characteristic curve (AUC), area under the precision-recall curve (AUPRC), clinical utility curves, and calibration plots. The interpretability of the model was enhanced through SHAP value analysis, followed by its deployment as a clinical decision support web application. The study included 715 patients from Center One and 224 from Center Two, identifying six key biomarkers as the core components of the SINTI model. Multivariable analysis confirmed that SINTI, clinical N stage, clinical T stage, and tumor size are independent predictors of early recurrence. The XGBoost algorithm exhibited robust discrimination during training cohort cross-validation, achieving a mean AUC of 0.860 (SD ± 0.021) and demonstrating consistent performance across validation datasets, with an internal AUC of 0.842 and an external AUC of 0.840. SHAP value interpretation revealed monotonic relationships between predictor variables and recurrence risk, with SINTI accounting for 36.1% of the total predictive weight. For clinical implementation, we deployed the optimized model as a web-based decision support tool, which can be accessed at https://p7toqbsdfbhlahdrugj4ra.streamlit.app/ . This interpretable AI framework demonstrates the potential to bridge data-driven modeling and clinical decision support, offering a transparent, potentially deployable solution for post-NCRT recurrence risk prediction following further prospective validation.

The monitoring of daily life in nursing home residents generates diverse and heterogeneous sources of information. Artificial Intelligence (AI) is increasingly used to predict a wide range of outcomes in both research and clinical practice, including mortality and cognitive impairment (CI). A key challenge is determining which information sources (IS) provide the most accurate predictions. In this work, we present an integrative AI-based framework that combines harmonized temporal modeling, Bayesian hyperparameter optimization, XGBoost, and explainable AI (SHAP) to predict CI in nursing home residents using 13 years of heterogeneous longitudinal data from 2,608 individuals. Our approach enables interpretable predictions of CI-related clinical scales such as the Mini-Mental State Examination (MMSE), the Global Deterioration Scale (GDS), and the Barthel Scale while revealing the relative contributions of diverse IS, including clinical metrics and activity records. Using a nested 5 × 3 cross-validation scheme with patient-level grouping and temporal blocking, the Bayesian-optimized XGBoost regressors achieved robust predictive performance, with MSE values of 2.12 (MMSE), 0.47 (GDS), and 4.55 (Barthel) when using only Clinical Variables, and further improvements when integrating all information sources (MMSE: 1.85; GDS: 0.42; Barthel: 4.30). The MMSE severity classifier achieved a macro-averaged AUC of 0.89 (95% CI: 0.87-0.91), with the highest F1-scores in the Normal (0.80) and Severe (0.86) impairment categories. Clinical Variables consistently emerged as the most informative source across regression and classification tasks. Overall, this integrative framework enhances CI prediction from heterogeneous long-term care data while providing interpretable insights that may support more personalized and data-informed care strategies.

Abstract Experimental characterization of brain white matter (BWM) using Magnetic Resonance Elastography (MRE), Diffusion Tensor Imaging (DTI), and numerical modeling is expensive, time-consuming, and constrained by computational limitations and model approximations. To address the scarcity of high-fidelity data, this study develops a machine learning (ML) workflow to predict single-frequency viscoelastic properties, specifically the homogenized storage modulus, of BWM. The dataset originates from a sensitivity study conducted in house where BWM was modeled as a 2D triphasic composite of axons, myelin, and glial matrix. The triphasic unidirectional composite only considers 2D mechanics and diffusion in the transverse plane (perpendicular to axonal direction). Microstructural properties such as fiber volume fraction, intrinsic phase moduli, and axonal geometry were used as features for the ML model. Ensemble of regression and decision tree-based models, coupled with hyperparameter optimization, were explored, with model interpretation performed using SHAP analysis. Decision trees yielded the best predictive performance, with SHAP highlighting the importance of glial moduli and fiber volume fraction. This ML framework offers a surrogate to expensive in vivo characterization, provides insight into BWM mechanical dependencies, and can serve as a foundation for future inverse models aimed at understanding aging, dementia, and traumatic brain injury mechanisms in neuroimaging studies

Multi scale deep residual network for single image super-resolution using mean-std normalization and Bayesian hyperparameter optimization

This study proposes a novel, domain-specific optimization framework for the Stable Diffusion XL (SDXL) model, addressing the critical challenges of structural consistency and aesthetic fidelity in AI-assisted interior design. Unlike generic applications of diffusion models, this research introduces a systematic pipeline integrating automated semantic cleaning with a rigorous hyperparameter optimization strategy. A high-quality, annotated dataset was constructed using a semi-automated YOLO-based filtering process to minimize noise. Furthermore, we established an empirically validated training protocol—combining optimal Dropout rates, L1/L2 regularization, and dynamic learning rates—specifically tuned to preserve the geometric constraints of interior spaces. Experimental results demonstrate that this optimized framework significantly outperforms baseline models, achieving superior Fréchet Inception Distance (FID), Structural Similarity Index (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS) scores, alongside robust CLIP Semantic Alignment. Furthermore, a systematic ablation study confirms that while domain-specific data provides the foundation, our semantic cleaning pipeline and structural regularization are critical for achieving high geometric fidelity, reducing FID by 51.1% compared to the baseline. The study contributes a technically robust methodology for adapting large-scale diffusion models to the specialized requirements of spatial design.

Accurate photovoltaic (PV) power forecasting under extreme weather conditions remains challenging due to the non-stationary and multi-modal nature of meteorological influences. This study proposes a novel four-stage learning framework integrating signal decomposition, hyperparameter optimization, temporal dependency learning, and residual compensation to enhance forecasting resilience during El Niño–Southern Oscillation (ENSO) climate transitions. The framework employs CEEMDAN for fluctuation mode decoupling, TOC for global hyperparameter optimization, Transformer model for spatiotemporal dependency learning, and EEMD-GRU for error correction. Experimental validation utilized a comprehensive dataset from Australia’s Yulara power station comprising 104,269 samples at 5 min resolution throughout 2024, covering a complete ENSO transition period. Compared against baseline Transformer model and CNN-BiLSTM models, the proposed framework achieved nRMSE of 1.08%, 7.04%, and 2.81% under sunny, rainy, and sandstorm conditions, respectively, with corresponding R2 values of 0.99981, 0.99782, and 0.99947. Cross-year validation (2023 to 2025) demonstrated maintained performance with nRMSE ranging from 4.68% to 15.88% across different temporal splits. The framework’s modular architecture enables targeted handling of distinct physical processes governing different weather regimes, providing a structured approach for climate-resilient PV forecasting that maintains 2.56% energy consistency error while adapting to rapid meteorological shifts.

Plastic recycling represents an essential element of strategies aimed at lowering global carbon emissions while supporting a circular plastics economy. However, the effectiveness of current plastic sorting systems remains limited by data scarcity, spectral variability, and the complexity of real world waste streams. This study introduces a CNN-based polymer classification framework that integrates physics-informed spectral simulation, adaptive data augmentation, and Bayesian hyperparameter optimization to enable robust classification under data limited conditions. Our framework combines near-infrared (NIR) spectral data from technical scale measurements with synthetically generated spectra. With only 100 measured spectra per polymer, the proposed framework achieves average balanced accuracies of 0.9739 in multi-target polymer classification tasks. By using technical scale spectral data, this study bridges the gap between laboratory model development and real sorting conditions.

Machine learning based prediction of vertical ground motion duration using hyperparameter optimization

AutoSGRL: Automated framework construction for self-supervised graph representation learning.

This study compares machine learning and econometric approaches for forecasting agricultural export values in volatile global markets, examining predictive accuracy and economic interpretability trade-offs. Monthly data from January 2014 to December 2023 were analyzed using five models: Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), Hybrid ANN-LSTM, Ordinary Least Squares (OLS), and Autoregressive Distributed Lag (ARDL). Key predictors included durian, mangosteen, and longan export values/volumes, plus China's GDP. Performance evaluation used MAE, RMSE, MAPE, and R² metrics with systematic hyperparameter optimization through grid search and 5-fold cross-validation. ANN achieved the highest absolute accuracy (MAE: 1,684,667,401.55; RMSE: 2,602,671,952.28), while Hybrid ANN-LSTM delivered superior relative accuracy (MAPE: 1.58%). ARDL demonstrated exceptional explanatory power (R²=0.83) for structural economic relationships. China's GDP emerged as the strongest determinant across all models. Longan export value showed contrasting effects between approaches, positive in machine learning models versus negative in econometric models, reflecting different paradigmatic interpretations of market substitution dynamics. This research introduces the first comprehensive comparative framework integrating advanced hybrid neural networks with traditional econometric methods for multi-commodity agricultural forecasting, addressing cross-commodity substitution effects previously unexplored while offering complementary perspectives for both predictive accuracy and economic policy interpretation.

A comparative study of computer vision models for oral cancer detection from oral photographs.

Automated detection of asymmetrical udders in dairy goats using a camera and deep-learning model YOLOv12.

Resource-adaptive successive doubling for hyperparameter optimization with large datasets on high-performance computing systems

This study aims to address the challenges in the diagnosis of oral cancer by proposing a novel computer-aided diagnostic framework that leverages advanced deep learning (DL) and optimization techniques to enhance early detection and improve patient outcomes. In the framework proposed, the histopathological images are subjected to a preprocessing technique, and then, the images are fed directly to the NASNet-Large model for the extraction of high-level discriminative texture features. The resultant vectors obtained from the features extracted act as input to the search space of Archimedes Optimization Algorithm that carries out dimensionality reduction and optimal hyperparameter tuning simultaneously. The optimized feature subset is fed to the final classifier, namely the Stacked Sparse Denoising Autoencoder that learns robust latent representations. The findings demonstrate that the proposed approach achieves superior performance, achieving an accuracy of 95.38%, a precision of 95.15%, a sensitivity of 91.78%, a specificity of 91.85%, and an F1-score of 93.72%. These findings underscore the potential of the SSDA-AOA framework as an effective tool for the early detection and precise classification of oral cancer, paving the way for improved patient outcomes through timely intervention. This innovative approach may significantly enhance patient outcomes by facilitating earlier diagnosis and treatment, addressing the urgent need for more reliable diagnostic tools in oncology.

Unsupervised learning of spatially varying regularization for diffeomorphic image registration.

Postoperative pain is a frequent clinical concern following endodontic treatment. This study aimed to develop and validate supervised machine learning models to predict the occurrence of postoperative pain in cases of irreversible pulpitis. A prospective sample of 354 patients aged 18 to 60 years undergoing standardised endodontic treatment was analysed. In the original randomised clinical trials from which the data were derived, each patient had only one eligible tooth included. Clinical variables included postoperative pain at 24 and 72 h, treated tooth group, occlusal reduction, photobiomodulation therapy, use of non-steroidal anti-inflammatory drugs (NSAIDs), sex and age. Eight supervised machine learning algorithms were trained to predict pain occurrence, including Logistic Regression, Support Vector Machine, Gradient Boosting, Random Forest, Decision Tree, K-Nearest Neighbours, AdaBoost and Multilayer Perceptron. The dataset was divided into training (70%) and testing (30%) sets using stratified sampling. Class imbalance in the training set, characterised by a lower proportion of cases with moderate or severe pain, was addressed using the Synthetic Minority Oversampling Technique. Hyperparameters were optimised through grid search combined with stratified five-fold cross validation. Model performance was evaluated using the area under the curve (AUC), accuracy, precision, recall and F1-score, with 95% confidence intervals estimated by bootstrapping. The predictive models achieved good discrimination of pain outcomes. Logistic Regression showed the best test performance at 24 h (AUC 0.74 [95% CI: 0.61 to 0.85], precision 0.81 [95% CI: 0.73 to 0.88]). At 72 h, the Support Vector Machine achieved the highest performance (AUC 0.81 [95% CI: 0.69 to 0.92], precision 0.88 [95% CI: 0.79 to 0.94]). Age and sex emerged as the most influential predictors across models. Supervised machine learning models demonstrated promising performance for predicting postoperative pain following endodontic treatment. Logistic Regression and Support Vector Machine algorithms presented the most consistent results, supporting their potential clinical application for personalised pain management.