Matching messy typed addresses to the right property
FLAP links free-text addresses to the UK's official property database with 99.2% adjusted accuracy and is released as open-source software.
The question
Health and social care records hold addresses typed as free text, with typos, abbreviations and local shorthand. To study care-home residents, housing moves or the geography of care, they must be matched to the official property record (the UPRN).
What we did
FLAP pairs a machine-learning classifier, which judges whether a typed address and an official record match, with a fuzzy alignment step that copes with typos and reordered words. We tested it on the addresses of 771,588 people in NHS Tayside and NHS Fife.
What we found
Adjusted matching accuracy was 0.992 on a test set of 3,876 addresses. It coped with typos, alternative formats and partly wrong information, and lets users set a confidence threshold or return the top candidates.
Why it matters
Open, validated tools for address linkage open the door to geographic health research. The code is on GitHub.