Machine Learning-Based Models for Predicting the Depth of Concrete Penetration

In the domain of concrete penetration, test data are often limited in quantity and unevenly distributed, which leads to poor accuracy of the machine learning-based model for predicting the depth of concrete penetration. This paper aims to improve the accuracy of the model within the constraints of limited penetration test data. In this paper, based on collecting a large amount of penetration test data, the penetration data was extended by data augmentation methods such as linear interpolation and adding Gaussian noise. The genetic algorithm and greedy algorithm were used to optimize four common machine learning models’ hyperparameters: multilayer perceptron (MLP), radial basis neural network (RBF), support vector regression (SVR), and extreme gradient boosting tree (XGBoost). The results show that using linear interpolation and adding Gaussian noise can effectively alleviate the problems of insufficient data and uneven data distribution. The average error of MLP, RBF, and XGBoost decreases by 2.7%, 3%, and 0.8%, respectively, after using data augmentation. In addition, the average error of the optimal machine learning-based method is 8.4%, and the global accuracy of this model is better than the commonly used empirical formulas. The machine learning-based model can effectively predict the depth of concrete penetration and meet engineering applications’ requirements.