Slashdot Mirror
False Positives, Few Matches Plague 'No-Fly' List
lindner writes "According to a recent article in the San Francisco Chronicle, the United States No-Fly List uses a soundex algorithm to match names. Designed 'to quickly summon passenger names or to catch deal-hunting passengers making duplicate bookings.' The system has only managed to rack up a slew of false-positives, including everyone matching soundex ("J. Adams") at one point in time. The problem has gotten so bad that there is now a "Fly List" for chronically misidentified passengers."
That algorithm is so fundamentally broken as to be practically useless for anything but as an aid in simple searches. Why anyone would use soundex in a mission critical application designed to positively identify individuals is beyond me. What, was the 'No Fly' database written by 1st year comp sci major or something? Sheesh.
Hey, wanna make a good living in these uncertain economic times? Come up with a better alternative and propose it to DARPA. There is actually some very cool research going on that is funded by DARPA in terms of biometrics and database centric comp sci.
Visit Jonesblog and say hello.
I think this has to do with the process of:
(1) Register for flight with Airline X.
(2) While registered for flight, search for better deal on similar flight with Airline Y.
(3) Cancel flight with Airline X.
Or I could be way off.
Here's a good run down on soundex and ten problems with it.
www.bannination.com Two things float to the top he
The metaphone algorithm addresses many of the shortcomings of soundex... why are they not using it?
what's wrong with passengers hunting for the best deal?
Story: to catch deal-hunting passengers making duplicate bookings.
This means that people were using different travel agencies to reserve lots of seats on the same flight, and then simply going with the agency who quoted the lowest price. This means that the seats end up being empty, and the rest of us end up paying for them. Sorry, but even the "rich elite" can no longer make duplicate reservations. Almost all airlines now use software to automatically cancel duplicate bookings.
Whenever the offence inspires less horror than the punishment, the rigour of penal law is obliged to give way...
This is unbelievable. Why reinvent the wheel, while there are at least 3 countries that have implemented similar restrictions and tracking systems succesfully for more than 10 years now (England, Israel, and Germany - that I know of).
This sounds like the work of some consultants with no idea of what they are faced with and fresh out of collage where they have analyzed a couple of algorithms... sorry - I have had way too much of these running around the office lately
If for once someone would just poke his head out and instead of trying to find a solution to an age old problem, look and see how others are handling it, we (taxpayers) would all be much more content (and safe...).
Just my 2c.
I guess all US people would screem for such a "threat to your privacy" ;)
But at least in here in sweden basically anytime you book a flight you give your ID number (similar to a social security #)
Two benefits:
1. Name is just a courtesy, so doesnt really matter for security if somebody get my name wrong.
2. On checkin, it must be _you_ not somebody with a similar name
Of course IDs can be stolen or forged, but that is a problem regardles of how you ID your self.
quote from the article: Scheduled for deployment in Spring 2004, CAPPS II will require airline ticket buyers to give more identifying information -- full name, birth date, home phone number and address. This information will be run against private credit-rating and government watch list databases to "verify you are who you say you are," Rosenker said
.] Now they're going to be checking that every time I fly?
What in the hell? I was under the impression that having a credit check actually hurts your credit history [as in, you shouldn't have too many credit checks in
I always hated flying, now I'm starting to hate air lines...
--- d'oh
I think the deal hunting they are refering to is the practice of buying 2 return tickets and then using the first set to get to your destination and the second to get back. This way you can get the cheap tickets that span weekend stays but instead travel on a, say, Tuesday and come home on Wednesday of the same week.
Even though you bought two complete return fares, it's still cheaper most of the time that buying one return ticket during a business week. They are trying to fleese the business traveller to subsidize the vacation traveller. I think you probably get both sets of airmiles from buying 2 sets as well
" That algorithm is so fundamentally broken as to be practically useless ..."
The problem is deeper than the algorithm. Soundex is quite useful when you are not quite sure of the spelling. For example, when someone gives their name over the phone.
Basic statistics is the issue. If there are 1,000 terrorists in the US and you have a procedure that is 99% accurate, then the test will nail 2,500,000 people. So you get 2499/2500 rate of false positives - 99.9+% false positives. Quickly, people will start to treat the alerts as a sick joke.
Any time you are looking for something rare, you are going to get a lot of false positives.
The same problem occurs with medical tests for rare diseases. Most people who come up positive on cancer screening tests do have have cancer. Cancer is common over a lifetime but rare at any given time in a given individual.
Face recognition technology has the same problem, looking for criminals in crowds.
John gilmore is suing for the right to travel anonymously(sp).
From the website:
He does so "because he believes persons have a right to travel by air without the government requiring that they relinquish their anonymity. No security threat is as important as the threat to American society caused by erosion of the right to travel, the right to be free from unreasonable searches, and the right to exercise First Amendment rights anonymously."
Check out the FAQ's, which are well written and explain the other reasons - including being subject to secret laws - he is opposing this.
HIV Crosses Species Barrier... into Muppets
From their Webpage:
I applied to this company recently for a software developer position, but I never heard back from them. I'm surprised soundex, which has been around for what, 40 years?, is actually used anymore.
You are in a maze of twisty little passages, all alike.
Obershelp/Ratcliffe works well for finding the best match in small sets, but on a large DB, it breaks down completely unless you are willing to dedicate significant time to the search. While better alternatives to Soundex definetely exist, Obershelp/Radcliffe is not one that should be used in large databases.
That's it. I'm no longer part of Team Sanity.
> If you think that anyone is ever again going to be able to hijack an airplane armed with a box-cutter - I think you're sorely mistaken.
> They would be savagely torn limb-from-limb by 300 homicidal passengers before the pilot even knew there was a potential hijacker
> onboard.
Which is essentially what happened during the recent attempted hijacking in Australia - passengers and flight attendants quickly overpowered the would-be hijacker. Unless you can stuff a plane with 10+% hijackers, it's not likely to work again any time soon.
Oddly, I noticed it the other day on a "terms of use" disclaimer for United. It went something like "you aren't allowed to play with our systems to get a better fare."
...say what their cost is?
The two tricks are "hidden cities" and "double-booking." For a hidden city, you book a flight through where you really want to go, get off the plane, but have a remaining segment. Since you had to make a stop, that ticket is cheaper.
The airlines really have a flawed pricing scheme. Maybe it made sense 15 years ago, but they really need to re-consider based on
The system relies on a false premise. Terrorists don't have "careers" anymore. If you were planning a terrorist attack, you could easily find 20 guys with no records whose names appear on no lists in any form.
These need not actaully be their real identities. Remember that 7 of those accused of being involved in 911 were proven to have used stolen identities.