Comparative study on machine learning algorithms for predicting compressive strength of high-strength concrete

Compressive strength is an important physical parameter for evaluating the mechanical properties of high-strength concrete (HSC). The compressive strength acquired under different fit ratios is different. To precisely forecast the compressive strength of HSC under different mix ratios, seven algorithms including linear regression, KNN, RF, DT, SVM, MLP and extreme gradient ascent were adopted on the compressive strength data of 324 samples. A comparative study was carried out on the compressive strength of HSC obtained from the selected 5 input variables (coarse aggregate, cement, water, fine aggregate and superplasticizer), and MSE and R^2 were used to evaluate the models. Finally, the prediction precision of different algorithms was compared by Taylor diagram. The calculation results of different algorithms show that the mean square error of the extreme gradient ascending method on the test set is 1.026 and the determining coefficient is 0.989, and its predictive performance is optimal. Three different kernel functions are used in SVR algorithm. The prediction accuracy of different kernel functions in the SVR is verified. The mean square error of the Gaussian kernel function on the test set is 1.732, The determinant coefficient is 0.981 and its predictive performance is better than that of the Sigmoid kernel function and the polynomial kernel function.