Abstract:
This thesis investigates the dynamics of crowdfunding success in Kickstarter’s technology category
using a blend of network analysis and machine learning. By sourcing a comprehensive
dataset from Webrobots.io[20], which includes detailed monthly data dumps of Kickstarter
projects, this study focuses on U.S.! (U.S.!)-based technology projects to uncover patterns and
factors contributing to successful crowdfunding outcomes.
The methodology encompasses rigorous data preprocessing, advanced feature engineering, and
graph-based modeling to represent relationships between projects, creators, and subcategories.
Network analysis techniques, including centrality measures and community detection, identify
influential projects and creators and uncover clusters with similar characteristics. A RandomForest
machine learning model, integrating project-specific metrics and network-derived features,
predicts project success with high accuracy.
Findings reveal significant patterns in project features, creator influence, and community structures
that impact crowdfunding success. The predictive model serves as a practical tool for
guiding project development and marketing strategies. This thesis enhances the understanding
of crowdfunding platforms through the integration of network science and predictive analytics,
offering insights for creators, backers, and platform moderators. The enriched dataset and
methodologies are made publicly accessible to encourage further research and practical applications
in crowdfunding.
