Predicting the donor journey using machine learning
LE3 .A278 2021
Master of Science
The fundamental question for fundraisers is what action they should take next to maximize their probability of receiving donations. One approach is to build a machine learning model based on the constituent’s past activities and useit to predict the next best action. To implement this, we used Long Short-term Memory Networks (LSTMs) which were able to learn from the time-series data and predict donations within $25 for one charity. Expanding upon this work, we tried different experiments using LSTMs, Bi-directional LSTMs, Convolutional Neural Network Long Short-term Memory (CNNLSTM), and Convolutional Neural Networks (CNNs). Furthermore, we experimented using data sets with different window sizes of past actions, including features that describe constituents data, and anonymously merging the data sets. Bi-Directional LSTMs and CNNLSTMs performed significantly better compared to LSTMs, while CNNs dropped the mean absolute error (MAE) for one charity to $16, giving the charities an improved model to choose actions and email parameters that should in turn improve donations.
The author grants permission to the University Librarian at Acadia University to reproduce, loan or distribute copies of my thesis in microform, paper or electronic formats on a non-profit basis. The author retains the copyright of the thesis.