Machine learning-based compressive strength prediction for concrete: An adaptive boosting approach

In this paper, an intelligent approach based on the machine learning technique is proposed for predicting the compressive strength of concrete. This approach employs the adaptive boosting algorithm to construct a strong learner by integrating several weak learners, which can find the mapping between the input data and output data. The weak learner whose predicting error is small will have a larger weight in the entire system, thus the overall accuracy of the strong learner will be enhanced. A total of 1030 sets of concrete compressive strength tests is collected to train and test the learners, in which the concrete mixture components (e.g., coarse/fine aggregates, cement, water, additive, etc.) and the curing time are set as the input data while the compressive strength value is set as the output data. The proposed approach is validated through a 10-fold cross validation method, and reaches an average accuracy of over 95% in sense of determination coefficient. In addition, a new dataset of 103 samples for concrete compressive strength is also adopted to demonstrate the generalization power of the proposed mode. The proposed approach is also compared to some other individual machine learning techniques that are already applied in this field, e.g., artificial neural network (ANN) and support vector machine (SVM), and shows superior advantages over these methods. Finally, the influence of some key factors in the adaptive boosting approach is also investigated, e.g., the amount of training data, the choice of weak learner, and the influence of the sensitivity and number of the input parameters. It is shown that using 80% of the total data for training can obtain acceptable prediction results and decision tree is the best choice for the weak learner in the boosting framework. Also, the importances of different input variables are obtained based on the sensitivity analysis results.