Jul 2026· Current Problems in Research· Vol 2, pp. 1-19· 0 citations· 39 references
TL;DR
A high-fidelity hybrid predictive framework that synergizes advanced machine learning architectures with nature-inspired metaheuristic optimizers that presents a highly reliable, physically explainable, and cost-effective alternative to ubiquitous physical testing, advancing the paradigm of digital geotechnical engineering.
Abstract
The precise estimation of spun pile efficiency parameters through High Strain Dynamic Testing (HSDT) is critical for structural safety and foundational integrity; however, it is frequently constrained by significant economic, temporal, and logistical limitations that restrict physical testing to a minor fraction of installed piles. To overcome these prohibitive barriers, this study proposes a high-fidelity hybrid predictive framework that synergizes advanced machine learning architectures with nature-inspired metaheuristic optimizers. Three distinct predictive models: Adaptive Neuro-Fuzzy Inference System (ANFIS), Support Vector Regression (SVR), and Regression Tree (RT) were formulated. To resolve the inherent limitations of suboptimal convergence and hyperparameter trapping in standard configurations, the structural parameters of these models were optimized using the Sparrow Search Algorithm (SSA), Whale Optimization Algorithm (WOA), and Manta Ray Foraging Optimization (MRFO). Utilizing a high-quality, rigorously validated dataset of 150 Pile Driving Analyzer (PDA) records, the models were trained to forecast the maximum case method capacity (RMX) and maximum compressive force (FMX). To ensure model robustness and completely eliminate small-sample bias, a rigorous 10-fold bootstrapping cross-validation protocol was implemented, alongside a SHapley Additive exPlanations (SHAP) sensitivity analysis and Wilcoxon signed-rank testing for non-parametric statistical validation. The comparative benchmarking confirms the absolute superiority of the SSA-ANFIS framework, which achieved unprecedented predictive precision with a perfect correlation (R2=1.000) and minimal error profiles for both RMX (RMSE=0.123) and FMX (RMSE=0.506), statistically outperforming baseline models such as eXtreme Gradient Boosting (XGBoost) and the empirical Danish Driving Formula. For practical field deployment, the optimized architecture was embedded into a MATLAB-based Intelligent Geotechnical Decision Support System (IGDSS) and subjected to independent validation on a geologically distinct site, verifying its exceptional generalization capabilities. The integration of swarm intelligence with neuro-fuzzy logic presents a highly reliable, physically explainable, and cost-effective alternative to ubiquitous physical testing, advancing the paradigm of digital geotechnical engineering.
Predicting the flexural capacity of reinforced ultra-high-performance concrete (UHPC) beams remains challenging due to the nonlinear interactions among longitudinal reinforcement, sectional geometry, material strength, and fiber-related parameters. This study proposes an explainable metaheuristic-enhanced CatBoost framework in which hyperparameter optimization is systematically compared across four strategies: grid search, particle swarm optimization, grey wolf optimization (GWO), and whale optimization algorithm. A curated experimental database comprising 232 beam tests with 15 input variables was assembled from multiple independent studies. Among the four developed models, GWO-CatBoost achieved the best generalization performance in the single-run evaluation, with test R2 = 0.975 and MAE = 6.072 kN·m, MAPE = 9.876%, and RMSE = 8.462 kN·m, substantially outperforming simplified mechanics-based benchmark formulations derived from NF P18-710 and ACI 318–19. Repeated-seed analysis across 50 random seeds further confirmed GWO-CatBoost as the most robust model, yielding the highest mean test R2 (0.9564 ± 0.0179) and the lowest RMSE (12.81 ± 3.08 kN·m), with the smallest variability among all candidates. SHAP-based interpretation revealed that longitudinal reinforcement area is the dominant predictor (52.0%), followed by beam height (13.2%), steel yield strength (7.8%), compressive strength (6.8%), and beam width (6.4%). At the same time, fiber-related parameters exhibit comparatively minor contributions. Partial dependence analysis further confirmed the physical consistency of the learned relationships. The proposed GWO-CatBoost framework provides an accurate, robust, and interpretable data-driven tool for predicting the flexural capacity of reinforced UHPC beams, offering practical guidance for metaheuristic-based optimization in structural engineering applications.
Accurate prediction of tunnel vault displacement and reliable assessment of deformation risk are essential for tunnel safety management under complex geological conditions. This study develops an integrated data-driven framework combining machine-learning prediction, metaheuristic hyperparameter optimization, statistical model comparison, and uncertainty-informed relative deformation-severity classification. Four baseline models, namely BP, SVM, LSTM, and Peephole-LSTM, were combined with GA, WOA, and PSO to construct 12 optimized models, resulting in 16 candidate models. Monitoring observations were chronologically divided into training, validation, and test sets at a ratio of 70%:15%:15%, and normalization parameters were calculated exclusively from the training set to prevent data leakage. Model robustness was further evaluated using rolling validation, the Friedman test, and post hoc Wilcoxon signed-rank tests with Holm correction. PSO-Peephole-LSTM achieved the best test-set performance, with an MAE of 0.058 mm, an RMSE of 0.074 mm, and an R2 of 0.884. PSO-Peephole-LSTM also achieved the lowest mean RMSE and average rank across the rolling validation windows, with statistically significant improvements over the competing models after Holm correction. Finally, 2000 Latin hypercube samples were generated to propagate prediction uncertainty and establish a five-level, project-specific relative deformation-severity classification framework. The integrated workflow provides a systematic connection between tunnel monitoring, displacement forecasting, uncertainty characterization, and uncertainty-informed monitoring and decision support.
Blast-induced flyrock is one of the most critical hazards in surface mining and quarrying, posing significant risks to occupational safety, nearby structures, and the environment. Accurate prediction of flyrock distance is therefore essential for safe blast design and effective risk management. In this study, Random Forest (RF), Extra Trees (ET), and Support Vector Regression (SVR) models were developed to predict flyrock distance, and their hyperparameters were optimized using the Secretary Bird Optimization Algorithm (SBOA) and the Spider-Tailed Horned Viper Optimizer (STHVO). Prior to optimization, six cross-validation strategies were compared using GridSearchCV to identify the most appropriate strategy for each model. The final models were evaluated on an independent test dataset using the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), variance accounted for (VAF), and Nash–Sutcliffe efficiency (NSE). Model interpretability was investigated using SHapley Additive exPlanations (SHAP). The results showed that appropriate cross-validation and metaheuristic hyperparameter optimization improved predictive performance, with the ET–STHVO model achieving the best overall results. SHAP analysis identified B/D, PF, H/B, RBS, and U/B as the most influential predictors. The proposed framework provides an accurate, robust, and interpretable decision-support tool for safer blasting operations.
Recycled powder mortar (RPM) represents a sustainable cementitious material whose compressive strength (CS) is governed by complex nonlinear interactions among material and mixture parameters. This study proposes an explainable, metaheuristic-optimized ensemble machine learning (ML) framework for accurate prediction and interpretation of CS of RPM. A dataset comprising 204 observations was employed, with recycled powder type (kind), particle size, water/binder (W/B), and mass replacement ratio (MRR) considered as input variables. Extra Trees (ET), Gradient Boosting (GB), and Histogram-Based Gradient Boosting (HGB) were integrated within bagging, voting, and stacking architectures, while hyperparameters were optimized using genetic algorithm (GA), particle swarm optimization (PSO), differential evolution (DE), and grey wolf optimizer (GWO). The dataset was partitioned into 80% training and 20% testing subsets, with model robustness further evaluated through shuffled 10-fold cross-validation. Predictive performance was quantified using R², RMSE, MAE, and MAPE. The GA-optimized ET–GB–HGB stacking model demonstrated superior predictive performance, achieving R² values of 0.949, 0.966, and 0.947, with corresponding RMSE values of 2.149, 1.931, and 2.129 MPa for the training, testing, and 10-fold cross validation datasets, respectively. SHAP and permutation importance consistently established the hierarchy MRR > particle size > kind > W/B, with MRR accounting for more than half of global SHAP importance. Sobol sensitivity and SHAP interaction analyses further confirmed MRR as the dominant factor and particle size–MRR as the strongest coupled interaction. Response-surface analysis additionally elucidated their nonlinear effects on CS. Overall, the proposed framework establishes a robust, interpretable, and data-driven methodology for accurate CS prediction and performance-informed design of sustainable RPM mixtures.
