Correspondence

摘要

Berk et al. (2009) have presented an approach that they have themselves developed: 'statistical learning by random forests' (SLRF) for forecasting rare events that are difficult to predict. However, they unfairly stigmatize logistic regression when used for this purpose. Their objection is that very few subjects are identified by logistic regression as more likely than not to be cases (posterior probability greater than 50%); and they imagine that nothing can be done to improve this.