Optimization of Data-Driven Ensemble Models Using Firefly Algorithm for Enhanced 3D Printed Concrete Strength Prediction (2025-09)¶

Three-dimensional concrete printing (3DCP) has emerged as a transformative construction technology. However, optimizing mix design for 3D-printed fiber-reinforced concrete (3DP-FRC) remains a multifactorial challenge due to the interplay of time-dependent rheology, printability constraints, and heterogeneous material compositions. This study develops six ensemble machine learning (ML) models-Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGB), and their Firefly Algorithm (FA)-optimized counterparts (FADT, FA-RF, FA-XGB)-to predict the compressive strength (CS) of 3DP-FRC. A dataset comprising 17 features of 299 experimental samples was used. Evaluation metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2) were employed. The FA-XGB model achieved the highest predictive accuracy with R2=98.87%,RMSE=4.7, and MAE =3.31 on the test set. FA optimization led to RMSE and MAE reductions of 19.44 % and 27.83 % for the XGB model, respectively. SHapley Additive explanations (SHAP) analysis identified water-to-binder ratio, OPC, superplasticizer, silica fume, and load direction as key influential features. Modified Taylor diagrams further validated the superiority of the FA-XGB model. The findings underscore the potential of integrating metaheuristic optimization with ensemble learning to produce robust, scalable models for 3DP-FRC mix design. The hybrid FA-XGB framework offers a reliable, data-driven approach for performance-oriented 3DCP applications.