Unlocking success in tech reward crowdfunding: A hybrid probit-machine learning approach with SHAP-driven feature analysis

This study examines the key factors associated with success in reward-based crowdfunding for technology projects by integrating traditional econometric analysis with explainable machine learning techniques. Drawing upon foundational theories such as Social Capital Theory and Information Signaling Theory, as well as empirical literature on crowdfunding, we select 14 key variables to evaluate their association with the likelihood of campaign success. Utilizing a meticulously manually compiled sample of 1,051 technology projects from Kickstarter (2009–2023) across all continents, we employ two methodological approaches, namely a Probit model to identify statistically significant factors and four tree-based machine learning algorithms (Decision Tree, Random Forest, XGBoost, and LightGBM) for predictive modeling and feature importance analysis. Our findings indicate that ten variables are significantly correlated with reward-based crowdfunding success within the Probit model, corroborating existing studies while extending insights to a global, technology-specific context. Among the machine learning models, XGBoost demonstrated superior performance in classification accuracy, as assessed by four standard metrics. Using the Mean SHAP (SHapley Additive exPlanations) technique, this study ranks success predictors, explains Probit and XGBoost differences, and exposes complex nonlinear predictors of success likelihood unseen by linear specifications. This dual-method approach not only affirms the robustness of our findings but also clarifies the complex, model-specific predictors. These insights identify key factors associated with campaign success within the crowdfunding ecosystem. This information can help campaign creators understand potential drivers of outcomes and may help backers in making more informed decisions by highlighting features correlated with successful projects. By aligning stakeholder incentives with empirically validated success factors, this study contributes to the development of more efficient and sustainable reward-based crowdfunding markets.