Regency EF
Using Machine Learning to Model Cancellation in Leadership Training Programs (304084)
*Sarah J. Pearsall, Center for Creative Leadership
*Philip Turk, Western Data Analytics, LLC
Keywords: machine learning, target encoding, group lasso logistic regression, classification, F1-score, confusion matrix
The Center for Creative Leadership (CCL) is a nonprofit global provider of leadership training and research. Leadership development programs at CCL are an essential piece of its business portfolio. These programs are susceptible to cancellations by prospective participants without warning, resulting in substantial revenue loss. The goals of this study were to discover why people are canceling and predict who will cancel in the future. To achieve this, we used a variety of machine learning techniques (e.g., group lasso logistic regression) in the software R to model participant cancellation based on a large number of predictors and factors. Using a classification decision rule that both maximizes precision and sensitivity while minimizing a cost function, our model is highly predictive of cancellation for a holdout data set. Based on important variables identified in model fitting and predictions from model deployment, a data-driven business strategy will allow CCL to stem and recover financial losses.