Using Supervised Learning Algorithms to Predict Discontinued Operations in Nonprofit Organizations

ABSTRACT Machine learning has been widely used to predict bankruptcy in for-profit companies. However, prediction of the discontinued operation of nonprofit organizations has rarely been studied. Although discontinued operation of for-profits is primarily due to financial issues, in nonprofits it can sometimes also be attributed to either completion of their mission or a change in needs of those they serve. This study employs machine learning algorithms to predict the discontinued operation of nonprofit organizations by focusing on six factors. Five groups of financial indicators for nonprofits are used to compare prediction performance of Logistic Regression, MultiBoost, Random Forest, Multilayer Perceptron, Support Vector Machine, and Bayesian Network routines. When a combination of all financial indicators is simultaneously used, Random Forest provides the highest accuracy of the procedures examined. Data Availability: Data for this study were obtained from the Internal Revenue Service (IRS) via Amazon Web Services. Recent data are available from IRS website. JEL Classifications: M41; M48; M49.