A crowdfunding prediction model: a data-driven approach

Abstract

The crowdfunding platform has been regarded as an important driver of creativities in that it can help innovators easily get financial support from the public to turn their ideas into reality. As crowdfunding environment develop dynamic, the outcome of a crowdfunding project become unreliable and is significantly affected by multiple factors. Due to the uncertainty of the outcome of fundraising cases, the development of crowdfunding prediction become an important issue and increasingly attracts much attention from innovators. In the prior research, some centered on the exploration of critically successful determinants of a crowdfunding project, whereas some focus on the development of a prediction model for a crowdfunding project by improving prediction accuracy. Although the result obtained by the prior research is remarkable, limited valuable information or suggestions contributing to project design can be derived from the research result. Accordingly, the research is motivated to propose a crowdfunding prediction model, named CPMCDM. By employing the techniques of text mining and classification algorithm, the model is proposed to extract knowledge beneficial for successfully predicting the crowdfunding outcome from a dataset containing 28,159 crowdfunding projects on Kickstarter. By applying 30-70% strategy, 261 classification rules were generated from the training dataset (70% of the collected dataset). The remaining 30% of the collected dataset was used for model testing and prediction accuracy was 87.19%. For ensuring the feasibility and applicability of the proposed model, CPMCDM was verified by specialists with crowdfunding relevant experiences based on the criteria including reasonableness of the generated rules and usefulness of the model. The proposed model is expected not only to contribute to the prediction of crowdfunding outcome, but also to assist in better campaign design for crowdfunding projects in the future.