Predicting University Donors: A Comparative Study of Classification Machine Learning Models

This study aimed to contribute to the fields of university fundraising and machine learning by applying four binary classification models to a large, novel dataset from a regional university foundation. The four models used were Logistic Regression (LR), Artificial Neural Network (ANN), Extreme Gradient Boost (XGB), and Support Vector Machine (SVM). Hyperparameter tuning was used to improve model performance, and a stacked ensemble method was employed. The results indicated that combining hyperparameter tuning with a probability threshold of 0.3 improved the model’s ability to correctly predict donors. Additionally, the ANN and the XGB were the optimal models in terms of recall and training time, while also maintaining acceptable accuracy, precision, and area under the curve. These results introduce ANN and XGB as viable models for binary classification in this context, ultimately contributing to the field and encouraging further research.