Skip to main navigation Skip to search Skip to main content

Explainable Hybridized Ensemble Model with a Metaheuristics Algorithm for the Prediction of Compressive Strength of Ultrahigh-Performance Concrete

  • Sanjog Chhetri Sapkota
  • , Sagar Shrestha
  • , Moinul Haq*
  • , Ajay Panthi
  • , Ajad Shrestha
  • , Sagar Sapkota
  • , Mohammed A. Al-Osta
  • , Tushar Bansal
  • *Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Ultrahigh-performance concrete (UHPC), composed of supplementary cementitious material, exhibits enhanced mechanical strength and durability, which increases its applicability in newer construction or the refurbishing of existing structures to prolong their strength and durability. Integrating industrial waste, including silica fume (SF), fly ash, and slags, helps mitigate environmental problems by reducing the need for cementitious materials. Several factors influence the development of UHPC, including temperature (T), relative humidity (RH), and concrete age. This study used extreme gradient boosting (XGB) as a primary model, and metaheuristics algorithms such as teaching-learning-based optimization (TLBO), biogeography optimization (BBO), jellyfish optimization (JOA), and krill herd optimization (KH) for the optimization of hyperparameters. This study used 10-fold cross-validation for all model combinations, which resulted in the optimal performance of R2>0.97 for testing sets when using XGB-KH. Similarly, other models such as XGB-JOA, XGB-BBO, and XGB-TLBO also had promising performance for the testing sets, with R2>0.96. Furthermore, for the model's explainability, Shapley additive explanation (SHAP) analysis was used in the study by employing the best-performing XGB-KH model. The study found that age, fiber, and silica fume are the most influential features when developing UHPC. The SHAP dependence plots confirm that cement content of 600-900 kg/m3 and SF content of 200-300 kg/m3 result in better compressive strength (CS). Likewise, with the increase in RH, there is an increase in CS, whereas increasing temperatures influence CS negatively. Furthermore, the experimental data sets with 0.96 in R2 validated the superior performance of the model. The relationship insights of every employed feature helps to carry out comprehensive assessments and establish robust processes for following sustainable construction practices.

Original languageEnglish
Article number04026023
JournalJournal of Structural Design and Construction Practice
Volume31
Issue number3
DOIs
StatePublished - 1 Aug 2026

Bibliographical note

Publisher Copyright:
© 2026 American Society of Civil Engineers.

Keywords

  • Ensemble learning
  • Industrial waste
  • Jellyfish optimization (JOA)
  • Krill herd optimization (KH)
  • Metaheuristics algorithm
  • Ultrahigh-performance concrete (UHPC)

ASJC Scopus subject areas

  • Arts and Humanities (miscellaneous)

Fingerprint

Dive into the research topics of 'Explainable Hybridized Ensemble Model with a Metaheuristics Algorithm for the Prediction of Compressive Strength of Ultrahigh-Performance Concrete'. Together they form a unique fingerprint.

Cite this