posteriorBalance: Posterior probability adjustment.

In traversc/trqwe: Performance oriented statistical metrics and utility functions.

Description

Adjusts the posterior probability of a classifier based on unbalanced datasets. In classification model where the negative data is randomly under-sampled and all the positive data is used, the adjustment factor (beta) is p(s=1|-) = p(+)/p(-). I.e., the probability that a negative datapoint is selected in the classifier. beta ~ N+/N-.

Usage

posteriorBalance(probs, beta = NULL, Nplus, Nminus)

Arguments

probs	The original posterior probability.
beta	The adjustment factor.
Nplus	The number of positive examples in the real data. Only used if beta is NULL.
Nminus	The number of negative examples in the real data. Only used if beta is NULL.

Value

The adjusted posterior probability.

See Also

Dal Pozzolo, Andrea, et al. "Calibrating probability with undersampling for unbalanced classification." Computational Intelligence, 2015 IEEE Symposium Series on. IEEE, 2015.

traversc/trqwe documentation built on Dec. 4, 2020, 4:21 a.m.