Comparative Analysis of AI-Based Models for Compressive Strength Prediction of Self-Compacting Concrete

Accurately predicting the compressive strength of Self-Compacting Concrete (SCC) is essential for attaining sustainable and high-performance construction with little trial-and-error. This study conducts a comprehensive comparative analysis of various models, including Artificial Neural Networks (ANN), Levenberg-Marquardt (LM), Machine Learning (ML) models such as Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), LightGBM, and a Deep Learning (DL) model developed with Keras. A constant set of data was employed in the preparation of twenty SCC mixes. The typical variables that had been used in each mix included cement, fly ash, the ratio of water/ powder, aggregates, and superplasticizer. It examined real values of compressive strength in all mixes (20 mixes) in the lab before churning out prediction models with the results. In the cases when the models could not give clear predictions, the trend-based estimations were used to give artificial values to ensure consistency in the study. In an extremely elaborate comparison of the entire models, we are using performance indicators such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Coefficient of Determination (R2). It is also highly interesting to use XGBoost and DL, particularly Keras. The LM optimised ANN model achieved the most accurate R2 = 0.999, and the lowest MAE. The prediction reliability of ANN-II was also confirmed experimentally. The created figures and tables provide a visual and statistical analysis of all the models within the 20-mix dataset. Such a combination methodology, founded on artificial intelligence, makes the SCC mix design even more precise and economical and allows people to make environmentally friendly decisions because they may reduce the amount of waste materials and laboratory tests. The outcomes of this work provide engineers with an easy-to-use aid in bringing AI to concrete and their predictability.