Discovering important features for charitable organizations using machine learning

In charities and non-profit organizations, a fundamental question for fundraisers is how to determine potential donors. The answer to this question could help fundraisers eminently before starting any fundraising campaign. One of the approaches is building machine learning models using past activities of constituents as training data to predict constituents' donationb ehavior. To achieve this, we implemented machine learning models with decision trees (DT) and LASSO regression (LR) to explore different types of donor behavior using features of donors. We formulated a novel synthesized approach using DT and LR with three different educational charitable organizations' data. In this study, four types of donor behavior were framed into four typ es of experiments. This approach was able to discover some important features to predict donor behavior. One of the experiments was able to identify that the number of people related to a constituent in the charity database can influence donation behavior. We also aimed to identify similarities in important features across different types of donor behavior. Our analysis identifed similarities in features where two experiments have some similarities in important features but opposite effects in describing the donor behavior. This study has a unique approach and should help charities understand their donors in an enhanced way using technologies from machine learning. Furthermore, This research is a groundwork in the realm of predicting donor behavior using features and many more possibilities can b e uncovered from here in the area of non-profits.