Enhancing the participation of PPP social capital is of great significance to stimulate effective investment. Select 61 characteristic variables in four dimensions of the project itself, local government, market environment, and macroeconomics, build a prediction model of social capital PPP participation based on an ensemble learning algorithm, and explore the influence mechanism of different feature combinations on social capital participation in PPP. The research results show that: 1) Dragonfly algorithm effectively improving the accuracy of the prediction models with feature dimensionality reduction. 2)The prediction model based on ADASYN+XGBoost achieves the best performance under the combination of the characteristics of the project itself, local government and macroeconomics. 3)Among them, the project itself has the most significant impact on social capital participation in PPP, followed by local government; the addition of market environment will reduce the accuracy of the model under other characteristic combinations except the project itself; local governments should focus on attracting active investment from social capital by improving fiscal transparency, improving integrity, and optimizing public-private cooperation mechanisms.