Crowdfunding Success Prediction Using Machine Learning: A Comparative Study Based on Türkiye’s Campaigns

As crowdfunding is widely used in finance, researchers have been interested in developing predictive models that can accurately assess crowdfunding campaign success. The purpose of this study is to create a machine learning based decision support system for the determination of crowdfunding campaign success in Türkiye. The study used 24 different machine learning models, and a dataset of 1,628 campaigns collected from 2011 to 2021 with 38 parameters. Tree-based ensemble models (Gradient Boosting, AdaBoost, CatBoost) achieved the highest classification accuracy of 99.4%, and performed much better than traditional classifiers, thereby showing their appropriateness for prediction analytics on crowdfunding success prediction. Accuracy, precision, recall, F1 score, and confidence intervals were used as performance metrics. The proposed framework reveals which features create the most impact on crowdfunding success prediction and finds strong correlations among social media and funding-related features in the crowdfunding dataset, highlighting key predictors like support rate and collected amount while identifying redundant variables to enhance model efficiency.