Human-In-The-Loop Based Success Rate Prediction for Medical Crowdfunding

Medical crowdfunding serves as a pivotal means of donor-driven funding to assist individuals unable to afford medical expenses. However, challenges such as a low success rate and suboptimal fundraising performances have garnered significant attention from medical crowdfunding platforms. This study employs a comprehensive framework combining neural network and tree models, augmented by Human-In-The-Loop (HITL), to predict the success rates of medical crowdfunding campaigns and identify the crucial determinants of fundraising effectiveness. Our approach enhances model interpretability, offering insights into the prediction and inference processes, and incorporates human feedback at various stages of model training and testing. We apply the method to a structured dataset from a leading medical crowdfunding platform. The findings indicate that our method achieves accuracy of 94.9%, AUC value of 98.2%, recall rate of 86.4%, and F1 score of 89.2% on the binary classification task. Further analysis reveals the primary factors influencing crowdfunding success to be the target amount and the duration of the fundraising campaign. These results prove the efficacy of incorporating HITL into the model development process, markedly enhancing performance and facilitating a deeper understanding of both the dataset and model predictions.