Kickstarter Predictor / The Importance of Plugging Leaky Data

May 9, 2018
Kickstarter Predictor / The Importance of Plugging Leaky Data
I chose a dataset that had a minimal amount of non-leaky data points to clearly illustrate the impact of having columns that leak the target being predicted through data points you wouldn't have prior to launching a Kickstarter. As you can see the non-leaky model still has a very good idea of what a live model looks like, and a reasonably good idea of if a project will fail, but performs only slightly over 50% for correctly guessing a success. For a more accurate Kickstarter predictor you'd likely want data points such as photos and text, as well as pledge perks offered.