Harnessing Deep Learning for Crowdfunding Success Prediction: A Comparative Analysis on Kickstarter Dataset

The rise of crowdfunding has transformed the landscape of fundraising for community projects, social initiatives, micro-enterprises, and startups, utilizing internet technology to connect donors with project creators worldwide. This research aims to evaluate the effectiveness of deep learning techniques in predicting the success of reward-based crowdfunding campaigns. We specifically applied Long Short-Term Memory (LSTM) models and a hybrid Gated Recurrent Units (GRU)-LSTM model, to conduct a critical analysis of the factors influencing crowdfunding success. Our Kickstarter project dataset, incorporating textual, numerical, and categorical features, forms the basis for this analysis. Results indicate that the Bidirectional LSTM model achieved the highest accuracy at 93%, while the Encoder-Decoder LSTM and hybrid GRU-LSTM models also demonstrated strong predictive performance, with accuracies of 92% and 91%, respectively. These findings offer valuable insights that can support backers in assessing the likelihood of project success, fostering more informed funding decisions, and enhancing crowdfunding outcomes.