Data Mining Algorithm results.

People,

Please excuse my ignorance, but I need some clarification on various
results specifically from Association and Sequential Pattern data
mining algorithms. Please be aware I am a blithering idiot when it
comes to anything other than the most basic understanding of
statistical tests. That said, I understand the concepts of support and
confidence as provided by each algorithm specifically. What I do not
understand however are the lift and p-values provided with the results.

Each of these tools provide a rule_body [A], rule_head [C] together
with support and lift. In the case of Association algorithm, confidence
is further provided. Each also provides p-value.

Could some kind soul please elaborate [in english] on how I would
interpret the lift and p-values and in conjunction with support /
confidence apply that to the results.

If for example the rules are mined over 1 or two day periods, then I
already calculate various values for these rules once mined. If support
is x, then I know the frequency of the rule based on the number of
transactions so I can figure days_rule_ocurred and I can also calculate
days_rule_head and days_rule_body.

These rules are returning p-values in the range of 19.19 - 37.76 [I do
not know if this is a percentage or not] with support of say 49.1803%,
confidence of 96.77% and lift of 1.64. Everything I have read points to
p-values of less than .05 by comparison.

I have no idea how to calculate a null hypothesis or if one is even
required, however I know that there is a 'null rule' that applies being
the support for the consequent [or rule_head]. If for example you are
mining over a period of 100 days and the rule head occurs 60 times in a
particular fashion then my null rule for that consequent is 60%. Does
this have anything at all to do with the null hypothesis ?

Once you have these rules [and there are a great many 000's of them]
how does one determine the most statistically relevant or predictive
ones from those that are not. Are there any further tests one can do on
such rules to further gain insight ? Chisq ofr example, if there is
then what data is required to perform such a test.

One last question, during mining we can see itemsets being generated
that also have support, p-value and lift [no confidence of course]. How
does one interpret those values to the final values that make up the
actual rule ?

Many thanks for any assistance you may be able to offer.

P

.

Relevant Pages

  • Re: AES with SslStream
    ... I can't remember which version of Windows is supposed to get that support, ... my understanding is that the AES types are still ... Aes The Advanced Encryption Standard algorithm. ...
    (microsoft.public.dotnet.security)
  • US-NJ-Imaging Physicist
    ... data and image processing in an ISO 13485 environment with strict ... support transition to real-time C-code.· ... Prepare software and algorithm documentation as needed.· ... the FDA's Quality System Regulation and the company's ISO ...
    (sci.physics)
  • RE: Password encryption
    ... For password encryption, could you let me know more about your application ... Because we usually don't need to use symmetric algorithm to ... Microsoft Online Community Support ...
    (microsoft.public.dotnet.framework)
  • Why does Feyd contrast so purely, whenever Ramez summons the necessary cathedral very around?
    ... Otherwise the lift in Frank's stool might snap some physical ... I was marking counterparts to ... Some acceptable rushs shake Mustapha, ... Just trying according to a text in support of the market is too ...
    (sci.crypt)
  • Re: transfer from wheelchair to toilet and back
    ... then sit her down on the toilet. ... we do not have a lift of any kind. ... using her legs and my support for balance. ...
    (alt.support.mult-sclerosis)