gpdEst: Estimate Threshold from Pareto Distribution

In RecordLinkage: Record Linkage Functions for Linking and Deduplicating Data Sets

Estimate Threshold from Pareto Distribution

Description

Fits a Pareto distribution to the distribution of weights and calculates a quantile on the fitted model as classification threshold.

Usage

gpdEst(Wdata, thresh = -Inf, quantil = 0.95)

Arguments

Wdata	A numeric vector representing weights of record pairs.
thresh	Threshold for exceedances.
quantil	A real number between 0 and 1. The desired quantile.

Details

The weights that exceed thresh are fitted to a generalized Pareto distribution (GPD). The estimated parameters shape and scale are used to calculate a classification threshold by the formula

\mathit{thresh}+\frac{\mathit{scale}}{\mathit{shape}} ((\frac{n}{k}(1-\mathit{quantil}))^{-\mathit{shape}} -1)

where n is the total number of weights and k the number of exceedances.

Value

A real number representing the resulting classification threshold. It is assured that the threshold lies in a reasonable range.

Author(s)

Murat Sariyar

See Also

getParetoThreshold for user-level function

RecordLinkage documentation built on Aug. 8, 2025, 6:05 p.m.