An innovative Machine Learning model for predicting compressive strength of biobased concretes

Biobased concretes, which incorporate renewable and environmentally friendly components such as plant-based aggregates, offer a promising alternative to conventional materials. However, their widespread adoption is hindered by several challenges such as variability in raw materials, complex interactions between components, the lack of standardized methodologies, and requirement of advanced technics for characterizing and optimizing their mechanical and hygrothermal properties.
This study explores, for the first time, the innovative application of machine learning (ML) techniques to predict the compressive strength of biobased concretes. To address these issues, a dataset was compiled from numerous previous studies, encompassing more than 200 different formulations of biobased concretes and 24 variables representing components properties and formulation details. A Decision Tree model was developed as part of the machine learning approach to analyze the dataset and predict the compressive strength of experimentally tested biobased concrete. This model operates by splitting the data into hierarchical decision nodes based on the most significant variables, creating a transparent and interpretable representation of the relationships between input features and output predictions. To maintain computational simplicity, the Decision Tree model’s maximum depth was restricted to 6. Additionally, the minimum number of samples required to split an internal decision node was set at 10, while each leaf node was constrained to a minimum of 5 samples.
For the test set, the Decision Tree model achieved an R² value of 0.72, a Mean Absolute Error (MAE) of 0.49 MPa, and a Mean Squared Error (MSE) of 0.52 MPa. While its performance still less accurate than more advanced models like the Artificial Neural Network (ANN), the Decision Tree provided valuable insights into the dataset’s structure and the relative importance of different material properties; this was clearly demonstrated when testing the model on two experimentally tested concretes. In sum, this research offers a fresh perspective by leveraging machine learning to address the complexities of biobased concrete, providing solutions to material variability and performance prediction.