Predicting the fundraising performance of environmental crowdfunding projects: An interpretable machine learning approach

Abstract

Crowdfunding has become a pivotal fundraising method for environmental organizations. However, the fundraising performance of environmental crowdfunding projects remains subpar, prompting the need for improvements. Effectively addressing this challenge entails the precise prediction of each project's fundraising performance and a comprehensive understanding of the intricate correlations between various features and fundraising success. In response to these imperatives, this study introduces an interpretable framework meticulously designed for predicting the fundraising performance of environmental crowdfunding projects. This comprehensive framework integrates ten theoretically significant features to form the predictive model's feature set. It adopts a diverse array of eight algorithms for training and harnesses SHAP values and ALE plots for insightful post-hoc interpretation, thereby providing valuable insights into the nuanced roles played by these features. Validated on a dataset comprising 3,101 environmental crowdfunding projects from Tencent Charity, the proposed framework outperforms state-of-the-art methods, demonstrating an improvement of 5.9% in predictive performance. Furthermore, the post-hoc interpretation techniques accurately depict the roles of the features. This study carries substantial practical implications for environmental crowdfunding project creators to optimize project design, for crowdfunding platform administrators to enhance platform performance, and for local governments to improve regional environmental governance.