A Unified Framework for Random Forest Prediction Error Estimation

We introduce a unified framework for random forest prediction err or estimation based on a novel estimator of the conditional prediction error distribution function. Our framework enables immediate estimation of key parameters often of interest, inc luding conditional mean squared prediction errors, conditional biases, and conditional qu antiles, by a straightforward plugin routine. Our approach is particularly well-adapted for p rediction interval estimation, which has received less attention in the random forest lit erature despite its practical utility; we show via simulations that our proposed predictio n intervals are competitive with, and in some settings outperform, existing methods. T o establish theoretical grounding for our framework, we prove pointwise uniform consiste ncy of a more stringent version of our estimator of the conditional prediction error distrib ution. In addition to providing a suite of measures of prediction uncertainty, our gener al framework is applicable to many variants of the random forest algorithm. The estimator s introduced here are implemented in the R package forestError .