UNIFYING CONVENTIONAL AND LIGHTWEIGHT CONCRETE: AN EXPLAINABLE AI WEB APPLICATION FOR COMPRESSIVE STRENGTH PREDICTION

Abstract

Accurate prediction of concrete compressive strength is important for mixture development, quality control, and efficient selection of construction materials. Conventional strength determination requires specimen preparation, curing, and destructive testing, which can make extensive mixture evaluation time-consuming. This study develops an explainable machine-learning framework for predicting the compressive strength of plain and foamed concrete using Random Forest and Extreme Gradient Boosting (XGBoost). A database containing 1,373 observations was compiled, comprising 1,030 plain-concrete and 343 foamed-concrete mixtures. Separate models were developed for the two material classes using an 80:20 training-to-testing division, while five-fold cross-validation was used during model development and tuning. Model performance was evaluated using the coefficient of determination, mean absolute error, and root mean square error. For plain concrete, XGBoost achieved a testing coefficient of determination of 0.937, compared with 0.883 for Random Forest. For foamed concrete, the corresponding testing values were 0.563 and 0.561, respectively. Shapley Additive Explanations (SHAP) were subsequently used to interpret the contribution of mixture variables to model predictions. The results show that prediction of plain-concrete strength is substantially more reliable within the present database, whereas prediction of foamed-concrete strength remains constrained by material heterogeneity, limited sample representation, and the absence of direct pore-structure descriptors among the principal model inputs. Because plain and foamed concrete are frequently used together within the same modern construction project or structural element, the underlying motivation for the study is a single, hybrid web-based application that predicts compressive strength for both material types from one interface, rather than through separate, disconnected tools.