Combining charitable data to build better machine learning models
LE3 .A278 2022
2022
Lee, Greg
Acadia University
Master of Science
Masters
Computer Science
A charity is a not-for-profit organization created to provide help and raise money for those in need. Charities use several approaches to reach donors and raise money. One such approach is machine learning. Machine learning strategies are used to increase retention and donation amounts. Machine learning models use data such as demographic, donation, education and behavioral to predict who will donate. Machine learning models are typically learned on data from a particular charity and applied to test data from the same charity. We experimented with learning models on data combined from multiple similar sources and all the available data from different sources. We hypothesized that these additional sources help build a more accurate model since the algorithm can access more and similarly distributed data. In our research, we wanted to show that combining data from both related and unrelated sources allowed for more accurate and precise models to be learned across a variety of charitable sectors. However, combining data did not help in general; it was dependent on the machine learning algorithm. We built a machine learning model called a super model. This model was trained using different machine learning classifiers such as Na ̈ıve Bayes, Logistic Regression, K-Nearest Neighbour, Random Forest classifier and Artificial neural network classifier. All the experiments were performed using different methods to determine the best charity and prediction pair. We discovered that Random Forest classifier and K-Nearest Neighbour classifier performed well when trained on the combined data available from all sources.
The author retains copyright in this thesis. Any substantial copying or any other actions that exceed fair dealing or other exceptions in the Copyright Act require the permission of the author.
https://scholar.acadiau.ca/islandora/object/theses:3885