Akademik Veri Yönetim Sistemi
Neural Computing and Applications, cilt.37, sa.28, ss.23495-23511, 2025 (SCI-Expanded)
This study presents machine learning models to predict compressive strengths of 924 CEM I 42.5 R type Portland cements. Particularly the utilized machine learning algorithms are adaptive network-based fuzzy inference systems, Random Forest, Support Vector Machine, Extreme Gradient Boosting, Light Gradient Boosting and Categorical Boosting. For machine learning, collected data contained 15 input features that show the physical and chemical properties of the cements. The compressive strengths at 1, 2, 7 and 28 days were defined as the output parameters. Models for each hydration day were trained with 748 data points and tested with 176 data points. Then, compressive strength test results and machine learning predictions were compared using statistical methods such as R-squared, mean absolute percentage error and root-mean-square error. The results indicate that Gradient Boosting models, in particular, accurately predict compressive strength, demonstrating that it is possible to estimate compressive strength without mechanical tests. In our developed Gradient Boosting model, the RMSE accuracy exceeds 95%, further supporting its reliability. The developed machine learning models offer substantial savings in both time and cost for compressive strength estimation.