ABSTRACT
Random Forests (FR) is an ensemble method
for regression, classification and recently for survival analysis. This note
explores the connections between the voting scheme for prediction and the
proximities, generated by RF output. We state that a kernel type estimator,
based on the proximities, is equivalent to a weighted voting scheme for
prediction.
ABSTRACT
In this paper, an approximation for the
quantile of the distribution of the sample mean of independent and equally
distributed observations is derived from the inversion of the Lugannani-Rice formula. As the inversion involves unknown
quantities which depend on the cumulant generating
function (c.g.f) of the population, the empirical c.g.f will be used to estimate them; this will yield to an
explicit empirical formula for quantile approximation. Its accuracy
is inspected, both theoretically and numerically.