The analysis of the composition of alkali-excited concrete in the civil engineering industry is of great significance for material development. However, there is no machine learning method suitable for alkali excited concrete composition analysis. Therefore, in this study, we use the test results of alkali slag concrete specimens as research samples. The composition of broken specimen blocks was analyzed by developing an ensemble learning algorithm called MLP-RF. Finally, by comparing the prediction results of this experiment with the prediction results of other scholars and optimizing the results, it is found that the ensemble learning algorithm has high accuracy and certain universality. This provides a feasible method to analyze the composition of alkali-excited concrete using machine learning.