9.5 Classification

The last step in a record linkage process - after records have been compared and weight vectors have been calculated - is the classification of record pairs into links, non-links, or if this decision should be done by a human review, possible links.

The current version of Febrl contains the classical Fellegi and Sunter classifier [13], as well as a flexible classifier that allows various ways of calculation of a matching weight. They are discussed in the following two sections.

Note: Future version of Febrl will include improved classification methods, based on sophisticated machine learning techniques. In fact, one of the main aims of our future research will be the development of better classification methods with the ultimate aim to reduce - or even eliminate - the class of possible links (and therefore reducing the time consuming clerical review process). For more information please visit the Febrl web site or contact the authors.



Subsections