The compressive strength of geopolymer is one of key factors in evaluating whether geopolymer can replace cement as a new building material, but relying only on many tests to test its strength wastes resources and improves costs. To solve this problem, the data of calcium carbide slag excited coal gangue geopolymer collected through early experiments, different mixing ratios, water-binder ratios, and ages were used as input parameters and compressive strength was used as output results. The strength response prediction model—Gaussian process regression (GPR) model was constructed based on machine learning methods. The geopolymer strength of different mixing ratios and ages was predicted by using the model, then the influence curves of each component content, water-binder ratio and age on the strength were established and the reasons were explored. The results show that the GPR model can predict the strength of geopolymer well after fitting the sample data, and the error is in the range of (-0.001 93~+0.001 83). The strength prediction of geopolymer with unknown compressive strength is made by the trained model, and the influences of each input parameters (calcium carbide slag content, coal gangue content, water-binder ratio, and curing age) on the strength were analyzed through the prediction results. It is found that the strength is closely related to the above variables, among which the calcium carbide slag content, coal gangue content and curing age have more influence on the strength.