Prediction of soil liquefaction during earthquake is a crucial task to mitigate or avoid damage caused by liquefaction. The existing machine learning methods have achieved satisfactory prediction accuracy on specific datasets, but they are unable to perform well on other datasets. To overcome the limitation, a novel prediction method based on stacking strategy are proposed to evaluate earthquake-induced liquefaction potential of soil, which is composed of six base classifiers and secondary classifier. The hyperparameters are tuned by grid search algorithm and the AUC value under ten folds cross validation are utilized as the basis for obtain the optimal hyperparameters. The applicability of stacking model was verified using three widely used datasets. Six performance metrics are utilized to analyze and compare the performance of base classifiers and stacking model. The result indicates proposed model outperforms base classifier in all three datasets in terms of the metrics mentioned above. Furthermore, the proposed method underwent a comparative evaluation against other existing machine learning techniques, revealing that the prediction accuracy achieved by the proposed model surpasses that of the existing methods. Also, this study investigated the importance of input parameters so as to interpret the complicated relationship between liquefaction potential and input parameters.