• An ensemble machine learning framework with NSGA-II optimization was developed to predict mechanical, ductility, and sustainability of RWS connections. • XGBoost demonstrated the highest predictive accuracy across most evaluated outputs within the ensemble machine learning models used in the study. • SHAP analysis showed cross-sectional geometry and material stiffness as key features influencing capacity and ductility, enhancing interpretability. • An online interface was deployed on Hugging Face to explore Pareto-optimal designs balancing seismic performance and embodied carbon reduction. A gap in current predictive modelling approaches limits the ability to accurately assess the mechanical, durability performance and sustainability metrics of Reduced Web Section (RWS) connections. This paper addresses this gap by developing an ensemble machine learning (ML) framework combined with multi-objective optimisation, enabling the efficient prediction of seven key mechanical and ductility properties alongside total embodied carbon (EC) reduction. Three ensemble ML models—Extra Trees Regressor (ETR), Gradient Tree Boosting (GTBR), and Extreme Gradient Boosting (XGBoost)—were evaluated, with XGBoost demonstrating superior generalization across most outputs. Additionally, Shapley Additive Explanations (SHAP) analysis was conducted to identify the most influential design parameters, improving model interpretability. The multi-objective optimisation performed using NSGA-II, generated Pareto-optimal solutions, highlighting trade-offs between structural performance and sustainability considerations. The findings reveal that cross-sectional properties, material stiffness, and connection type significantly impact RWS performance, and optimising these parameters can lead to improved ductility, moment capacity, and reduced environmental impact. To enhance practical applicability, a user-friendly interface was developed and deployed via Hugging Face, allowing users to test the results, make predictions and retrieve optimal design parameters based on the nearest Pareto-optimal solutions. The results of this paper demonstrate that ensemble ML methods, coupled with optimisation and explainability tools, provide a robust framework for advancing RWS connection design, ensuring both seismic resilience and sustainability in structural engineering.

Physics-informed ensemble learning for hierarchical fault diagnosis in quadruped robots

Prediction of groundwater total nitrogen via an interpretable ensemble machine learning framework: Implications for groundwater diversion management in complex catchments.

Development and validation of stability prediction models for intracranial aneurysms based on ensemble learning algorithms.

Intelligent fault diagnosis in power systems using Random Forest and ensemble learning

Enhancing financial fraud detection with graph neural network and ensemble learning: insights from Related Party Transactions network

Mathematically grounded state of charge estimation via Gershgorin-based feature engineering and ensemble learning from electrochemical impedance spectroscopy data

Hydrothermal liquefaction (HTL) of biomass is a promising thermochemical pathway for sustainable biofuel production, yet accurately predicting the process energy demand remains challenging due to the nonlinear interplay between feedstock properties and operating parameters. This study aims to develop robust machine learning models to predict HTL energy consumption (MJ/t biomass) using a comprehensive dataset of 653 experimental records drawn from peer‑reviewed literature. Input features include elemental composition (C, H, N, S, O, ash), biochemical composition (protein, lipid, carbohydrate), and process conditions (temperature, reaction time, solid loading ratio). Seven algorithms, Decision Tree (DT), AdaBoost (AB), Random Forest (RF), K‑Nearest Neighbor (KNN), Ensemble Learning (EL), Convolutional Neural Network (CNN), and Multilayer Perceptron‑ANN (MLP‑ANN), were optimized through targeted hyperparameter tuning and evaluated via 5‑fold cross‑validation using R², MSE, and AARE%. RF achieved superior performance with a test R² of 0.936, low MSE (142,229), and minimal AARE% (3.19), followed closely by KNN. SHAP analysis revealed temperature as the overwhelmingly dominant predictor (impact magnitude ≈1150), with reaction time and protein content playing secondary but consistent roles. Positive SHAP values for high temperature and prolonged reaction times confirmed their contribution to higher energy demand, while protein‑rich feedstocks exhibited negative SHAP trends, reducing predicted requirements. Findings underscore the capacity of optimized tree‑based algorithms to deliver accurate, generalizable energy demand forecasts across diverse biomass types and process regimes. These models offer actionable insights for HTL operational optimization, reducing energy intensity and supporting scalable bioenergy deployment. • Modeling Energy Demand in Hydrothermal Liquefaction of Biomass. • temperature as the overwhelmingly dominant predictor. • Random Forest achieved the highest generalization capability.

A multi-criteria approach to risk-based sustainability performance scoring: Integrating environmental, structural, and green aspects with fuzzy systems and ensemble learning

Accurate and rapid capacity estimation is essential for efficient battery management in industrial settings particularly for cell grading, pack assembly, and second-life screening where throughput, cost, and energy efficiency are paramount. Conventional approaches require complete discharge cycles, leading to testing times of several hours per cell, which severely limits scalability and increases operational costs. To address this bottleneck, this paper proposes a fast capacity estimation method for battery capacity grading in the production process, which utilizes only the early-stage voltage measurements within the first 300-480 seconds of the initial discharge cycle during cell grading to accurately predict the cell’s nominal capacity, enabling reliable battery capacity grading within minutes instead of hours. Although real-world grading data from production lines are often inaccessible, this first-cycle setup serves as a well-controlled surrogate that replicates key aspects of factory-based capacity labeling. The method exploits early-voltage transients that encode degradation-sensitive electrochemical signatures such as lithium inventory loss and solid-electrolyte interphase (SEI) evolution arising from microscopic changes in charge-transfer resistance and ion transport dynamics. From this short window, we extract physically interpretable health indicators (HIs) that reflect underlying aging mechanisms. A nonlinear feature enhancement strategy is then applied to amplify subtle capacity-related patterns while suppressing manufacturing-induced variability. These engineered features feed into a Multi-Decision Ensemble Learning (MDEL) architecture, which adaptively fuses multiple regression pathways to improve robustness across diverse cell chemistries and aging stages. Evaluated on both in-lab cells, the public CALCE and MIT dataset spanning fresh to end-of-life capacity conditions, the proposed approach achieves a mean absolute error (MAE) of ≤0.0391 Ah (≤1.63% of nominal capacity), which is comparable to the methods with complete cycle data while reducing testing time by over 80%. This enables reliable capacity assessment in minutes rather than hours, offering a practical, scalable solution for high-throughput battery manufacturing, precise pack matching, and rapid second-life qualification. • Enables fast capacity estimation during battery grading with early discharge data. • Phase-aware segmentation captures capacity-sensitive features in initial discharge. • Nonlinear enhancement decouples health indicators for representation learning. • Multi-decision ensemble achieves high accuracy across our laboratory, CALCE, MIT datasets.

An ensemble machine learning framework for predicting the soil-water characteristic curve of compacted soils

Hybrid classification with ensemble representation learning for subject-independent EEG-based motor imagery

Assessment of flood susceptibility in Minab County, Iran, through the integration of topographic, climatic, and land-surface indices using ensemble machine learning models

Bi-directional long short-term memory based ensemble deep learning framework for non-linear steam turbine power forecasting: a biomass fuelled case study

Ensemble deep learning framework integrating deep image features and statistical descriptors for robust tumor diagnosis

Machine learning-based prediction of drug lactation risk: Bridging molecular features and breastfeeding safety.

Diabetic Retinopathy (DR) is a leading complication of prolonged diabetes, which poses a significant threat to vision and may lead to permanent blindness. Early identification and timely intervention are crucial to preventing disease progression. Traditionally, DR diagnosis relies on medical examination of retinal fundus images by expert ophthalmologists, which is time-consuming and resource intensive. However, deep learning techniques, particularly medical imaging, have demonstrated remarkable performance in the automated detection and classification of DR. This study proposes an ensemble-based deep learning framework using feature-level fusion stacking, which integrates four complementary convolutional neural networks named ReXInDen and three complementary convolutional neural networks named ReXDen for automated DR detection from retinal fundus images. These frameworks extract high-level features from each backbone, concatenate them into a unified representation, and classify using a feedforward neural network. Three datasets were utilized to validate the model including a region-specific dataset collected from Bangladeshi medical sources. The proposed ReXInDen model achieved accuracies of 98.27%, and 98.69% on Dataset 1 and Dataset 2, while ReXDen achieved the highest accuracy of 99.05% on Dataset 3. These results indicate a substantial improvement over individual models and demonstrate the potential of the ensemble approach to support early-stage DR detection. Moreover, these models show promise for integration into automated DR screening tools that can aid in reducing the global burden of diabetic vision loss. • Develop an ensemble method combining different convolutional models for medical image-based diagnosis. • Apply advanced image filtering techniques to improve retinal image clarity before analysis. • Merge features from multiple models and classify them using a feedforward neural network. • Validate the method using three datasets, including one from local healthcare providers. • Achieve 99.05 percent accuracy with the ensemble method on the combined dataset.

Identifying high-risk individuals for cardiovascular and all-cause mortality among individuals with cardiovascular-kidney-metabolic (CKM) syndrome stage 0-3 can guide the implementation of targeted interventions. This study aimed to evaluate the predictive value of plasma proteins for future cardiovascular and all-cause mortality. This study included 39,007 participants from the UK Biobank (UKB) with CKM stage 0-3 and available proteomic data. Associations between plasma proteins and future risks of cardiovascular and all-cause mortality were assessed using Cox proportional hazards models. Key proteins were identified through an ensemble machine learning approach integrating support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) algorithms. Subsequently, Cox models were applied to evaluate the incremental predictive value of these key proteins and their ability to enhance risk stratification for mortality outcomes. Furthermore, temporal trajectories of protein levels were examined in the years preceding death. During a median follow-up of 15.2years, 505 participants died from cardiovascular causes and 3368 from any cause. 56 and 269 out of 2911 plasma proteins were significantly associated with cardiovascular and all-cause mortality, respectively (Bonferroni-adjusted P<0.05). Incorporating seven and eight key proteins into conventional model significantly improved long-term predictive performance (C-statistics: 0.812 versus 0.782 for cardiovascular mortality; 0.772 versus 0.739 for all-cause mortality; both P<0.001), and also provided incremental predictive value for 5- and 10-year mortality risks. Notably, participants died during follow-up exhibited markedly elevated certain protein levels over a decade before deaths, with progressively increasing trajectories over time. Stratification based on optimal predicted risk thresholds further revealed distinct cumulative mortality risks across groups. In individuals with CKM stage 0-3, plasma proteins combined with traditional risk factors may predict future cardiovascular and all-cause mortality.

Ensemble machine learning and landsat observations reveal seasonal and spatial dynamics of water quality in a river-influenced estuarine system

Toward trustworthy web attack detection: An uncertainty-aware ensemble deep kernel learning model