Data Science Project: Kickstarter Forecast

Please review our reports and slides here, video here, and codes at Github.

Online crowd fundraising or crowdfunding has revolutionized the way we support creative arts and technological innovations. Unlike traditional ways of fundraising, an artist could pitch a new album via a crowdfunding platform, in the hope that thousands would give a small amount of monetary support, rather than seek significant amounts of investment from a few sources. 


On most of the crowdfunding sites (e.g. Kickstarter, Indiegogo), a creator is who initiated the idea and set up the fundraising process, while a backer is a person who contributed money to the project. Often, backers are promised some reward when the project is complete, for instance, an album made by the artist. In this scenario, a backer is not only an investor who might actively involve in the project promotion, but also a customer who expects the delivery of a product. Every project has a pledging goal set by the creators. There is an all-or-nothing principle: if the fundraising goal is not reached by the deadline, all the funds are returned to the investors, while all commitments are canceled. Such policy helps investors to avoid the risk of paying for nothing, but it also renders their expectation futile. Therefore, predicting the outcomes of crowdfunding projects is crucial for both creators and backers.

For our group project, we propose to predict the accomplishment of a crowdfunding campaign. It is not a simple divide between success and failure. The accomplishment of a crowdfunding campaign is measured by to what degree the actual fund exceed the expected goal.  A project campaign that received three times more than its goal is considered more accomplished than those that just meet the line. Previous researchers primarily examined crowdfunding through a simple criterion – success or fail to meet the goal. Their results have addressed the effects of project properties, such as category, duration and goal (Chen et al. 2015), as well as language used in the project campaign (Mitra et al. 2014, Aleyasen et al. 15). Further studies investigated the roles of social influence, including social media activity (Lu et al. 2015, Kamath et al. 2016), and creator trustworthiness (Rakesh et al. ). Building upon these existing knowledge, we are also curious about the effects of the broader social structural indicators, including the economic and income level of the birthplace of a project idea. Such influence is invisible but deep rooted. We want to build a baseline model with features mentioned above, and a new model to compare the prediction accuracy.

We primarily investigate the following research questions:

  • RQ1: How can the socio-economic factors, including the local disposable income and the currency used by the Kickstarter campaign predict the campaign performance?
  • RQ2: How can the campaign properties, such as the campaign goal, reward strategy and the campaign category predict the campaign performance?
  • RQ3: How can the social engagement of creators and backers, such as #updates, #comments, #Facebook friends,  predict the campaign performance?

The dataset will be Kickstarter project data, as well as the economic data of the cities.

Jiahui Wu

I’m currently a Ph.D. student at iSchool.profile-pic

Yuting Liao