Application of machine learning for predicting concrete strength: Ensembles versus instance‐based algorithms in WEKA platform

Abstract

This research work presents a comparative analysis of machine learning techniques for predicting the compressive strength of concrete, a critical parameter in civil engineering. The study evaluates instance‐based learning methods, such as locally weighted learning (LWL), K*, and IBk, against ensemble‐based methods like Bagging, Random Committee, and Ensemble Selection utilizing the WEKA software platform. The findings indicate that ensemble methods significantly outperform instance‐based methods in terms of prediction accuracy with improvements in correlation coefficients ranging from 6% to 10%. Ensemble methods, particularly Random Comm and Ensemble Selection, demonstrate superior performance with correlation coefficients of 0.9359 and 0.9583 respectively compared to the highest correlation coefficient of 0.8701 achieved by the instance‐based method IBk. The research also underscores the importance of data preprocessing and employs Spearman's rank correlation for statistical analysis. These insights contribute to the advancement of ML (machine learning) applications in the construction industry and offer a quantitative basis for the comparative strengths of different ML algorithms for predicting concrete compressive strength. This study underscores the engineering significance of reliable concrete‐strength prediction while systematically exploring ML‐based investigation, reporting enhanced predictive outcomes and illustrating their practical value in mix‐design optimization. The innovation lies in integrating rigorous statistical filtering with a comparative evaluation of ensemble and instance‐based learning, offering a unified framework for high‐accuracy concrete strength prediction. The models demonstrated strong predictive performance, with correlation coefficients exceeding 0.95 and RMSE reductions of over 40%, supporting their suitability for reliable compressive strength estimation.