Using machine learning approach towards successful crowdfunding prediction

Abstract

Crowdfunding is a concept that emerged due to difficulties in raising funds for community business projects, social activities, micro-enterprises, and start-ups conventionally. Crowdfunding uses internet technology as a bridge between the donor and the recipient of funds so that it can reach a wider range of donors. This study aims to compare the performance of machine learning approaches in predicting crowdfunding campaign success. Three machine learning algorithms were employed to predict crowdfunding campaign success, namely logistic regression, random forest, and extreme gradient boosting (XGBoost). The dataset used in this study contains data about all projects posted on Kickstarter from January 2020 to September 2022. To improve the prediction model's performance, experiments using principal component analysis (PCA) feature reduction and log transformation were conducted. The results show that the implementation of log transformation on the dataset can increase the prediction model's performance. Meanwhile, XGBoost algorithm performs better than linear regression and random forest.