Fueled by media coverage of massive success stories, Kickstarter has become a popular crowdfunding platform where creators can seek funds to turn their ideas into products. However, only a third of projects manage to reach their fundraising goals and little is known about the ingredients required to make a successful Kickstarer campaign. To solve this problem, we analyzed over 380,000 Kickstarter campaigns from 2009 to 2018 and ran a plethora of machine learning algorithms to predict whether a campaign will be a success or not. In general, accuracy was low across all models due to underfitting. Tree-based models performed better with XGBoost as the winner with an accuracy of 70.23% and recall of 43.64%. Our results indicate a need for more features by way of feature engineering or scraping more data to improve the performance of our models.
The full report of our findings can be found in project_writeup.pdf.
Jupyter notebook of our code is contained in Final_Project.ipynb.