getErrorMeasures-methods {RecordLinkage}

Calculate Error Measures

Description

Computes various error measures for the classification of a data set.

Details

Let \mathit{TP} be the number of correctly classified matches (true positives), \mathit{TN} the number of correctly classified non-matches (true negatives), \mathit{FP} and \mathit{FN} the number of misclassified non-matches and matches (false positives and false negatives). The calculated error measures are:

alpha error

\frac{\mathit{FN}}{\mathit{TP}+\mathit{FN}}

beta error

\frac{\mathit{FP}}{\mathit{TN}+\mathit{FP}}

accuracy

\frac{\mathit{TP}+\mathit{TN}}{\mathit{TP}+\mathit{TN}+\mathit{FP}+\mathit{FN}}

precision

\frac{\mathit{TP}}{\mathit{TP}+\mathit{FP}}

sensitivity

\frac{\mathit{TP}}{\mathit{TP}+\mathit{FN}}

specificity

\frac{\mathit{TN}}{\mathit{TN}+\mathit{FP}}

ppv

Positive predictive value:\frac{\mathit{TP}}{\mathit{TP}+\mathit{FP}}

npv

Negative predictive value:\frac{\mathit{TN}}{\mathit{TN}+\mathit{FN}}

Value

A list with components alpha, beta, accuracy, precision, sensitivity, specificity, ppv and npv, each a number in the range [0,1].

Methods

signature(object = "RecLinkResult")

Method for S3 result objects of class "RecLinkResult"

signature(object = "RLResult")

Method for S4 objects of class "RLResult", from classification of big data objects (see "RLBigData", "RLBigDataDedup", "RLBigDataLinkage")

A wrapper function errorMeasures(result) exists for compatibility with package version 0.2.

Note

Record pairs with unknown true matching status (e.g. due to missing values in the argument identity to RLBigDataDedup) and possible links are not counted, which can distort the values returned by this function.

Author(s)

Murat Sariyar, Andreas Borg