Predictive Modeling for Kickstarter Campaign Success

Context

Needed to forecast whether Kickstarter campaigns would succeed, with campaign features varying widely and prediction baseline showing low accuracy.

Problem

Campaign features varied widely and prediction baseline was low accuracy, limiting ability to advise campaign creators.

Approach

• Performed comprehensive preprocessing and data cleaning

• Implemented advanced feature engineering techniques

• Tested multiple ML models including logistic regression, random forest, and SVM

• Optimized model parameters through cross-validation

• Delivered final submission with contest-compliant format

Outcome

• Improved campaign prediction accuracy from 71% to 78%

• Delivered final submission with high accuracy metrics

• Identified key features driving campaign success

• Created reproducible R workflow for future predictions

Next Steps

• Expand feature set to include social media engagement metrics

• Explore ensemble techniques for higher accuracy

• Develop real-time prediction API for campaign creators