Ensemble Learning for Crowdfunding Dynamics: JingDong Crowdfunding Projects

Hu Min, Kaihan Wu, Minghao Tan, Junyan Lin, Yufan Zheng, Choujun Zhan

Abstract: As an emerging internet financing model with high efficiency, low cost, diversified returns, and a small investment, crowdfunding is sought after by entrepreneurs and investors. However, many crowdfunding projects are faced with the risk of low success rates and failure to reach the financing target within the specified period. Therefore, the prediction of crowdfunding project financing results and multi-model comparison are important ways to improve the project success rates and reduce market risk. First, we collected project data of JingDong (JD) crowdfunding platform for preprocessing and analyzed the characteristics of successful projects. Then, we use ensemble learning and traditional machine learning models to predict the daily amount of crowdfunding with grid search to obtain the optimal hyperparameters of each model. Several evaluation metrics are then employed to assess the performance of the model.The experimental results demonstrate that the Extra Tree Regression (ETR) ensemble model achieves the best prediction performance, with a coefficient of determination(\(R^{2}\)) of 90.1%, when forecasting the daily crowdfunding fundraising. Furthermore, the ensemble learning model showed significant advantages in other evaluation indicators, indicating its potential in forecasting the financing amount of crowdfunding projects.