Estimating compressive strength of CO ₂ incorporated concrete with data augmentation and explainable regression modeling

To mitigate the negative impacts of concrete production on the environment while simultaneously improving its structural performance, innovative approaches are required to meet the growing demand for environmentally friendly building materials. This study investigates the integration of regulated dosages of Carbon Dioxide (CO 2 ) during the concrete mixing process to improve compressive strength while contributing to carbon sequestration. Further, the study addresses the challenge of estimating concrete strength by utilizing machine learning algorithms. Moreover, this is accomplished only by utilizing limited experimental data consisting of only 270 rows. To make the Machine Learning (ML) model learn patterns well, Conditional Tabular Generative Adversarial Networks (CTGAN) and Tabular Variational Autoencoder (TVAE) are utilized to generate high-quality synthetic data, thereby significantly expanding the training dataset. This allows for avoiding the limitations associated with small datasets generated from experimentation. The data generated by CTGAN and TVAE are evaluated to ensure that the characteristics of the synthetic data are aligned with the original dataset. After data is compiled using combined real and filtered synthetic datasets, significant improvements in key performance metrics are achieved. Furthermore, Explainable Artificial Intelligence (XAI) techniques are incorporated to provide insights into the black-box models, improving their interpretability and trustworthiness. Results indicate that the Light Gradient Boosting Machine is the most effective for predicting compressive strength, achieving an R 2 value of 0.9872, MAE of 1.1847, and RMSE of 1.3833 on test data. The approach proposed demonstrates a robust framework for combining data augmentation and ML to estimate CO 2 incorporated concrete compressive strength while effectively addressing challenges related to data scarcity.