Advanced Visualization and Interpretable Machine Learning for Performance Prediction of Biochar-Modified Concrete

The integration of carbon-based materials in cementitious systems has been of interest as a method to improve both the sustainability and performance of concrete. This research presents a data driven methodology to evaluate the mechanical, workability, and durability performance of biochar modified concrete. The five machine learning models, TabPFN, TPOT, TabICL, Deep Kernel Regression, and Bayesian Neural Network, were trained to predict the performance of the concrete using the collected data. Results indicated high levels of predictive ability with TPOT having the highest accuracy (R2 = 0.89) of all models tested. Analysis further revealed that biochar content ranging from 1-2% would be most effective in balancing both the mechanical stability and the durability performance of the concrete. In general terms, the results of this research show the effectiveness of utilizing advanced multivariate visualization techniques and machine learning algorithms to assist designers in developing sustainable biochar modified concretes.