Compressive Strength Prediction of Recycled Aggregate Concrete Based on Different Machine Learning Algorithms

Abstract

The use of recycled concrete aggregate (RAC) in creating new concrete has gained significant attention for environmental and financial reasons. However, the compressive strength of the product concrete is hard to predict due to many variables. In this study, the compressive strength of recycled aggregate concrete was predicted using eight well-known machine learning algorithms, including support vector machine (SVM), artificial neural network (ANN), XGboost, Tree, Random Forest, Gradient Boosting, CatBoost, and AdaBoost. Every machine learning algorithm's general methodology entails gathering and analyzing input data, training the algorithm, testing the algorithm, and producing an output. A total of 419 data samples (experimental tests) were used in training and testing all the machine learning models. The results show that the best models for estimating RAC compressive strength are Neural Network, AdaBoost, and XGBoost. The other algorithms, random forest, gradient boosting, and Catboost, performed well in predicting the compressive strength of RAC, however, tree decision and SVM performed badly. The primary evaluation metrics used in this study were Mean Squared Error (MSE) and R-squared (R²), which helped determine the accuracy and reliability of the predictive models.

Keywords

• Compressive Strength, Machine learning, Neural Network, Recycled Aggregate Concrete