A Data Science Approach for Predicting Crowdfunding Success

Crowdfunding is important for backing innovative projects and new startup businesses. However, success in achieving the target fundraising is a big challenge, and it depends on many complex factors. This work uses data science to predict the success of crowdfunding pledges using a historical dataset that was scrapped from the Kickstarter website. The dataset was subject to intensive data wrangling, exploration, and engineering procedures. Three machine learning models were constructed in this study using: (1) Random Forests (RF), (3) K-Nearest Neighbor (KNN), and Support Vector Machine (SVM) algorithms. The models were trained using a separate portion representing two-thirds of the dataset, while the remaining third was used for evaluation. The KNN model achieved the best performance with a classification accuracy of 97.9% and an AUC of 98.3%. Random Forests was the second-best model, with a classification accuracy of 94.9% and an AUC of 98.9%. The Precision, Recall, F1, and AUC metrics also confirmed the validity of the reported results, while the confusion matrix and the calibration curve confirmed the robustness of the constructed models.