Predicting crowdfunding success with optimally weighted random forests

The social media version of fundraising, crowdfunding has, within a very short time, become an important and often critical element of the process that allows budding entrepreneurs to grow in todays market. While the underlying principle behind crowdfunding - to get multiple people to financially back an idea and collectively amass enough resources to make it happen - is simple enough in theory, it is much harder to actually implement. Kickstarter, the largest and most well-known crowdfunding website internationally, sees more than half of the uploaded projects rejected due to inadequate funds gathered. In this study, we take a closer look at the reasons for success and failure of these projects from the project description pages, where the creators attempt to persuade potential backers into supporting their respective creations. A dataset of over 26 thousand Kickstarter projects was created and used to understand the factors that most affect the chances of acquiring successful funding. In this study, we propose a novel algorithm for the RandomForests learning model in which we assign optimal weights to the individual classifiers and perform weighted majority voting on them using the 13 predictors that were found to be most suitable through implementation logic. These predictors will serve to be useful to both, a project creator as well as a potential backer, in order to devote their energy and resources judiciously. Finally, we tested our optimized learning model on the freshly acquired dataset and observed 94.289% accuracy, 94.5% precision, 94.3% recall and 94.3% F-measure value in success prediction. This research achieved the best accuracy and greatest amount of success in this field.