paul-jm/kickstarter_campaigns

Forecasting Campaign Success

Kickstarter Campaign Forecasting

Kickstarter would like to reduce its proportion of failed campaigns, i.e. campaigns for which the amount of money pledged by backers is inferior to the user's initial goal.

Data

The plublic dataset comes from Kickstarter - data/input/ks_dataset.xlsx.

Pipeline

Pipeline is detailed and commented pipeline.ipynb. Requirements are listed in requirements.txt. Several auxiliary functions are in functions/ for different steps of the code.

Outcome

The goal is to forecast the pledge_usd variable - the amount raised by each campaign by taking into account the campaign name, duration, month launched, the currency & country of origin, as well as the product's main category.

The prediction can be suggested as a goal to every user, in order for them to decide whether their original objective was too ambitious. By deploying an XGBoost model - parameters are not optimized as there is significant overfitting, assuming all campaigners used our prediction as their objective, we have reduced significantly the proportion of failed campaigns.