Prediction and Analysis of Success on Crowdfunding Projects

Kickstarter was established in New York, USA in April 2009. This is a crowdfunding platform for those who have ideas and need funding for projects. The information at the launch time of the project is used to train an optimal machine learning model to predict the success of the project. A lot of data is crawled on the kickstarter website and extracted as features that can be used for training. Logistic Regression(LR), SVM, MLP, AdaBoost, Random Forest(RF), XGBoost and GBDT models are trained and grid search is used to obtain the best hyperparameters. The experimental result shows that the GBDT model is better than other models in all indicators. And the accuracy rate reaches 70.2% based only on the information at the launch time.