An efficient record linkage scheme using graphical analysis for identifier error detection

Table 6 Classifier performance on an independent validation set of 25,000 complex clusters

Status	Predicted not bad	Predicted bad	Total	% predicted bad
Unknown	5623	2501	8124	30.7
Good	15975	901	16876	5.33
Bad	337	11698	12035	97.2

The logistic classifier derived to identify bad clusters (not bad refers to a single individual within a cluster, bad refers to more than one individual), shown in Table 4, was applied to a further random sample of 25,000 clusters obtained after initial record linkage. These were classified into 'good' 'unknown status' and 'bad' using rules, as described in Table 4 Legend and methods. The classifier performance on this validation set is shown.

ISSN: 1472-6947