What is the state of the art for matching identities using software
tools ?  For example, how do criminal justice systems match up people
within their systems without having exact social security numbers ? 
How do these databases match a person named "Achmed" and "Ahmed" ? 
Who are the people that offer this type of software and what are the
primary matching algorithms ?

Hello misterfine,

I will break the answer into two sections - the more general query
first and then the specific items you asked about.

On "identity matching", I will assume you mean how a system will
select one (or several) records for further processing. Generally
these systems are looking for a "good match" based on say 6 of 10
criteria - the more matches are better. Some criteria may be weighted
more than others as well. For the most part, a variety of methods are
used and I'll provide pointers to suppliers of this kind of software.

http://www.searchsoftware.com/Products/index.htm
  They have a pretty good family of products - both searching and
grouping as turnkey applications and as tools to build into a
customer's product.

https://www.nascio.org/awards/1998awards/AdminApps/westvirginia.cfm
  A description of how West Virginia linked up with federal systems
for DNA matching. More information generally means better matches.

http://www.crimtrac.gov.au/fingerprints.htm
  An interesting summary of how Australia automated the collection and
processing of fingerprints (and hand prints). With prints, you deal
with spatial information and have less likelyhood to get an exact
match. A person may be needed to confirm the match.

http://www.forensic-evidence.com/site/EVID/EL_DNAerror.html
  A look at errors in DNA matching (and what that means for forensic
analysis).

You can get a variety of sites with search phrases such as "match
identity police system", (or replace police system with justice,
legal, forensic analysis, ...).

On matching with social security numbers or with similar names
(typographic errors or homonyms), the methods used are typically
called "fuzzy" searching or matching. I will provide a few links to
sites that provide search methods including:

http://www.htdig.org/htfuzzy.html
  Ht://Dig is a free search tool, often delivered as a part of Linux
systems (e.g., Red Hat) and can do fuzzy searching with synonyms,
soundex, and a number of other methods.

http://citeseer.nj.nec.com/518489.html
  A very technical paper describing methods of matching images using
fuzzy techniques. It relates how to do some spatial relationships as
well as context based matching.

http://www.mrexcel.com/fuzzy/fuzzy.shtml
  A contest from a year ago or so that asked authors to do fuzzy
matching on strings in Visual Basic - with several code examples
listed. This will give you an idea of the complexity of such a search.

and so on. A search for "fuzzy search", "fuzzy match", or similar
phrases will give you plenty of sites to review.

Please let me know if you need any of this clarified.

  --Maniac

---

Request for Answer Clarification by misterfine-ga on 06 Aug 2002 09:25 PDT

I appreciate the thoroughness of the answer, but it didn't get to the
exact point I was looking for, perhaps because my questions was not
targeted enough.  (I'll pay for the answer, though).  Perhaps you also
know the answer to this :

A variety of systems contain information about people : credit
reporting, criminal records, etc.  MAny of them need to be linked so
that a researcher can traverse between them and develop a picture of
an individual, either to determine their credit worthiness or a
criminal pattern.  However, people will often use different social
security numbers to hide themselves, as well as different spellings in
their names (or different names).  A criminal, for example, caught in
a different state will give a false social security number so as to
not be identified as a fugitive in their home state.  Alternately, a
consumer's name might be spelled "Achmed" in one database and "Ahmed"
in a different one (without social security numbers).

How do credit reporting and criminal agencies match up their databases
and link people when their information does not exactly match ?  Who
builds this type of Identity (human identity) matching software ?

---

Clarification of Answer by maniac-ga on 06 Aug 2002 19:21 PDT

Hello misterfine,

This is response to the clarification request. I did a quick search on
google with both
  match identity of criminal from multiple databases
  match identify of criminal from multiple databases
and found a number of sites that describe the methods used to
integrate information from a number of databases to identify a
criminal. As I mentioned in the original answer (and described in
these other sites) a match does not have to be exact. To use a few
examples of criteria
 - social security number
 - drivers license number
 - automobile plates
 - fingerprints
 - facial features
 - height, weight, hair and skin color
 - credit cards
 - other identification
As described in 
  http://www.ppdonline.org/ppd2_tech_ibis.htm
the Philadelphia police provide a hand held unit that allows the cop
on the street to collect some of the data above, capture photos and
fingerprints for analysis and comparison with the suspected criminal.
It also looks up arrest warrants.

You might not have all the information that is provided. You might not
have all the data in the same format. Some of the data may be
incorrect. The software is written to do the comparison (generally
fuzzy matches), and return the "best matches". That doesn't mean that
every speeder you pull over will have a criminal record, but it will
give you information for the few you do (when you have enough data for
a partial match).

  --Maniac

The question was a little vague, but the answer hit what I was looking for.