As a Lamdaschool data science student, me along with other data science students was tasked to create a web app that would predict the success/failure of a Kickstarter campaign. It was a fun project to do and it felt great to create a fully functioning useful web app within just four days! Investing in a Kickstarter campaign is not risk free. If a campaign fails, the investors loose all the money. So, you have to look at a campaign to determine the possibility of the campaign being successful before putting your money on a campaign. Only if there was someone who has a great intuition of determining which one’s will succeed! Lucky for you our app can do exactly that.
We have taken the data of all the Kickstarter campaigns from a website named Web Robots Which contains all the information we needed to create a predictive model, that would facilitate our app with the power to predict the success/failure of a Kickstarter campaign based on some features of the given campaign.
The features we used to create our model was:
• Category
• Description length
• Campaign goal (in USD)
• Campaign length (days)
• Staff pick
Exploring the datasets, we found out that out of 209,445 Kickstarter campaigns:
• 121,651 were successful
• 73,683 failed
• 8,904 were cancelled
• 5,207 are live
After some more Exploratory data analysis, we Used “Random Forest Classifier” to create our predictive model which gave us around 75.3% test accuracy which we were pretty satisfied with considering how many different variables can affect the success of a campaign which makes creating a super accurate model impossible.
To use our predictive model, we developed a Flask app which allows you to put the features as input and spit out the probability of your campaign being successful.
After spending a good amount of time analyzing the data, creating model and app, we would suggest if you want to start a Kickstarter campaign, or you want to invest in one, make sure:
• Have a reasonable goal
• Don’t have too long campaign
• Have a good description about the campaign
Credits:
• Antonio Peterson - Project lead
• Jace Hambrick - Data Engineer
• Jashim Rashid - Data Engineer
• Luis Urene - Machine Learning Engineer
• Ekram Ahmed(Myself) - Machine Learning Engineer
Resources:
• Datasets
• App