Machine Learning and Non-Investment Crowdfunding Research: A Tutorial

Abstract

This study is intended for researchers (and doctoral students) interested in learning more on the use of machine learning methods in non-investment crowdfunding (i.e., reward- and donation-based). In particular, the study illustrates the insights that machine learning methods could provide on non-investment crowdfunding, for example, through data and information visualization, the ranking of features importance, and prediction assessment metrics. Specifically, I use four machine learning methods (gradient boosted decision trees, random forests, shallow neural networks, and support vector machines). As the literature shows, machine learning methods outperform classical regression models when the underlying relations are nonlinear. As such, the study offers some insights on the nonlinear relationships that could exist between the explanatory variables and the likelihood of success for art projects (e.g., threshold and Goldilocks effects). The study also offers some guidance to art project creators.