•  
  •  
 

Document Type

Original Article

Abstract

The production of conventional concrete has a significant impact on the environment, primarily due to the cement (C) component. To mitigate this impact, binders with reduced environmental footprints were used in concrete mixtures as a partial replacement for C, including powders of industrial byproducts. Waste marble powder (WMP) is one such alternative; however, its incorporation causes complicated interactions with cementitious materials, making the compressive strength (CS) prediction difficult. This study developed and evaluated four predictive models (multilinear regression (MLR), full quadratic regression (FQ), M5P tree, and artificial neural network (ANN)) using a dataset of 306 WMP-blended concrete mixes from 16 published studies. Statistical evaluation parameters (mean absolute error (MAE), root mean squared error (RMSE), coefficient of determination (R2), scatter index (SI), and objective function (OBJ)) were used to assess the performance of models. The ANN model achieved superior performance on the testing subset (R2=0.96, RMSE=2.95 MPa), while 10-fold cross-validation confirmed its predictive robustness, yielding mean and RMSE values of 0.93 0.04 and 3.54 1.17 MPa, respectively. The SHAP (SHapley Additive exPlanations) method of sensitivity analysis illustrated that C and curing time (t) were the main factors affecting CS, while WMP had a negative context-dependent effect at higher levels of replacement. The results indicated that ANN models can successfully capture the nonlinear behavior of WMP-modified concrete and can be used as a practical tool for optimizing concrete mix designs for sustainability.

Receive Date

15 Apr 2026

Revise Date

15 July 2026

Accept Date

25 July 2026

Publication Date

2026

Download

Share

COinS