Optimized Machine Learning Models for Estimating Soil Compaction Parameters From Index Properties
Advances in Civil Engineering, cilt.2026, sa.1, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 2026 Sayı: 1
- Basım Tarihi: 2026
- Doi Numarası: 10.1155/adce/9973351
- Dergi Adı: Advances in Civil Engineering
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, INSPEC, Directory of Open Access Journals, Middle East & Africa Database (ProQuest), Engineering Source (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: artificial neural network, giant Pacific octopus optimizer, maximum dry density, metaheuristic optimization, optimum moisture content, Proctor compaction, SHAP interpretability, soil index properties
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
Accurate estimation of maximum dry density (MDD) and optimum moisture content (OMC) is important for geotechnical design; however, laboratory compaction testing is often time- and resource-intensive. This study presents a metaheuristic-enhanced machine learning (ML) framework coupling the giant Pacific octopus optimizer (GPO) with artificial neural networks (ANNs), Lasso regression, and random forest (RF) models, using five routinely measured index properties (gravel content [G], sand content [S], fines content, liquid limit [LL], and plastic limit [PL]) as inputs. A dataset of 182 soil samples with diverse gradation and plasticity characteristics was used to develop and evaluate the proposed models using repeated 5-fold cross-validation, with an independent test set employed for additional verification. Among the developed models, GPO-ANN achieved the highest repeated cross-validation accuracy for MDD, with a mean R2 of 0.8500 ± 0.0424, while for OMC it attained a mean R2 of 0.8383 ± 0.0460, closely comparable to that of GPO-Lasso (0.8413 ± 0.0426). These results indicate that GPO-ANN provided the most consistent predictive performance among the three investigated GPO-based models across different data partitions. Cosine amplitude method (CAM)- and SHapley Additive exPlanation (SHAP)-based analyses assigned the highest importance to S and Gs for MDD prediction, whereas both gradation- and plasticity-related variables contributed to OMC prediction. These patterns were broadly consistent with expected geotechnical behavior. The proposed framework provides an interpretable approach for the preliminary estimation of MDD and OMC from routinely measured soil index properties. It may support early-stage geotechnical assessment and testing prioritization, while laboratory compaction testing remains necessary for project-specific design and verification.