To address the issue of insufficient overall performance of traditional machine learning algorithms in preterm birth prediction, an innovative preterm birth prediction method based on Stacking model was proposed. Firstly, during the data preprocessing stage, an under-sampling technique was applied to balance the distribution of positive and negative samples, and numerical differences between variables were eliminated through data standardization. Secondly, feature selection was carried out by carefully analyzing the correlations between features and assessing their importance scores. Then, in the construction of the Stacking model, the Pearson correlation coefficient was calculated among the prediction results of different machine learning algorithms, and this analysis was used to adjust both the type and number of base classifiers. Finally, a comprehensive evaluation of the preterm birth prediction method based on the Stacking model was conducted using multiple evaluation indicators, and compared and analyzed with existing methods to verify the effectiveness of the method. The results show that the proposed method achieves remarkable performance, with scores of 0.921 9 in AUC, 0.922 9 in Accuracy, 0.916 4 in F1 score, and 0.858 5 in Recall. These results significantly outperform the best performances of the 11 individual models used to build the Stacking model, and the overall performance is better than the existing research methods. The proposed method can effectively identify high-risk individuals for preterm birth in early pregnancy, providing strong support for early intervention in early pregnanly.