Online Crowdfunding Projects Performance Prediction

This paper is dedicated to Identification of determinants and predicting the success of crowdfunding projects placed on online intermediary platforms. The considered sample is narrowed down to video games related projects. Due to uniqueness and high specificity of required data, dataset was obtained using software which collects relevant information automatically directly from projects web pages. Final dataset contained information about 8520 unique videogames projects in 32 features of interest. To analyze patterns in the data both traditional econometrics toolbox (Logit and LASSO-regressions) and machine learning algorithms (generalized additive models, random forest, gradient boosting, support vector machine) were used. In addition, stacking procedure of constructing algorithms compositions which is popular in data analysis competitions was applied. The main result of this paper is not only in detection and interpretation of significant determinants of online crowdfunding project’s success but also in developing high performance predicting algorithms (87.95% accuracy was achieved by 8-fold cross-validation).