[IP] a comment on "Phones studied as attack detector"

[David Farber <dave@farber.net>]

Begin forwarded message:

From: Valdis.Kletnieks@vt.edu
Date: May 8, 2007 3:58:23 PM EDT
To: dave@farber.net
Cc: ip@v2.listbox.com
Subject: Re: [IP] a comment on "Phones studied as attack detector"

On Tue, 08 May 2007 11:21:52 EDT, David Farber said:
> Detectors for bio events are notoriously flaky. They see false
> posoitives in  the 3-4% for mass deployed units. Consider if the
> False Positive of the cell phone detector was even 1%. The noise
> would be overwhelming.

Even if they get the FP rate down to 0.01%, you still hita problem -
statisticians call it the 'base rate fallacy'.  Floyd Rumin discussed it
well here: http://www.lewrockwell.com/orig7/rudmin1.html in the  
context of
NSA data mining of phone calls and e-mail to look for terrorists.

Basically - you can't use data mining to find something very rare,  
because
the false positives *will* drown you unless you have an insanely good  
tool
to do the good/bad classification.

And even after all that, the terrorists can twist it to their  
advantage...

Consider a terrorist cell that learns how to game the system and cause
false positives at will - the 395th time they have to clear a major  
sports
stadium in the middle of a game, or close down the New York subway  
system,
they'll give up on it.  Of course, at that point, the terrorists have
a 2-for-1 special on the advantage:

1) They'll have a detailed understanding of exactly what the response  
time
and capabilities of responding units are.

2) The 396th time....




-------------------------------------------