Machine learning to predict the compressive strength of mortars with and without construction waste

ABSTRACT

The present work aimed to evaluate the performance of machine learning algorithms in predicting the compressive strength of mortars. The database was created through a bibliographic search of more than 50 references that were cataloged to contain data on mortar dosages with or without the addition of construction waste. The dataset used in the experiments underwent preprocessing, which included the integration of construction and demolition waste data and normalization. For normalization, the z-score technique was chosen. Then, the algorithms, linear regressions, decision trees, ensembles, and neural networks were used to predict compressive strength. The dataset was separated into 80% for training and validation and 20% for testing. The cross-validation was of the k-fold type with ten divisions in the training subset. Evaluating the performance of the models, the ensemble Gradient Boosting algorithm showed the best performance when compared to the others, reaching a value greater than 89% in the coefficient of determination. Finally, it is concluded that Machine Learning (ML) is a practical calculation tool for predicting the compressive strength of mortars. Furthermore, the artificial intelligence model was prototyped for the scientific and technical community use in a web version available through the Python Streamlit framework.

Keywords
Artificial Intelligence; Mortars; Waste Aggregate; Prediction; Construction and Demolition Waste Data