Hacker News Reader

Comment by galactus | Hacker News Reader

Pc86 knows the meaning of false positives and false negatives, you should read his original question again.

DanBC: "The article mentions a 90% detection success rate, but it doesn't go into rates of false positives (people who aren't scammers who are flagged as scammers) or false negatives;"

Then...

pc86: "Maybe I'm misunderstanding your question, but doesn't 90% detection imply a 10% false negative rate"

Then...

atwebb: explains difference between false positive and false negative.

Then...

pc86: "The question was about false negatives, not false positives. The parent's point was that he wanted to know the false negative rate when they found 90% of fraud attempts. I was asking if that 90% number meant that by definition there was a 10% false negative rate."

But... the parent's point was about false positives, not false negatives. In order to clear up the dialogue I put the definition of each, then re-asked the question which was not previously answered.

The question still stands: What are they going to do about people who are detected as fraudsters who are actually not trying to commit fraud?

Good lord.. The original comment was about both false negatives and false positives: "it doesn't go into rates of false positives [...] or false negatives"

Pc86 pointed that we actually know the rate for false negatives, if we assume that 90% success means that the system did not detect 10% of real frauds. So he was asking if he misunderstood the meaning of "90% success".

edit: typo

Honestly, I am struggling with your comments.

Since the false negative question was answered, why are you focusing on it? The false positive question is far more interesting as it will impact individuals, rather than the buyers of the service who will work to understand the false negatives and the risk implications.

To me, the false positives will be the same as someone having their identity stolen and not being able to get credit. This has wide reaching implications for those individuals. Its likely not an issue for them as a company, but I hope they are thinking about the implications so they are not harming individuals.

[added] I assume everyone knows what false positives, false negatives are, the point of my original comment was to redefine the thread since the false negative problem is addressed in the TC article.

atwebb's comment was not germane. He talked about both FPs and FNs when FPs had nothing to do with my question. If he had explained how a 90% detection rate does not necessarily mean there is a 10% false negative rate, then it would have been relevant.

DanBC said the article did not mention FPs or FNs. I agreed with the FP point so from here out lets ignore anything related to FPs. My question/point was that as far as I understood it that 90% detection rate by definition means there is a 10% rate of false negatives.

You and I view FPs and FNs differently.

FNs to me are not interesting because they will be studied by the service buyer so there is someone obviously interested in researching it and the market will make sure that FNs are low enough or the service will not make any money. FNs will automatically be solved. Asking about them doesn't even make sense unless you are a buyer.

FPs on the other hand are not important to anyone but potential 'victims' (too strong of a word, but still) and those victims do not have any way to negotiate their position in the process.

By forcing people to use the system by signing up service providers, the company is creating a situation where FPs are a very big deal for those that are flagged as potential fraudsters.

I could care less about FNs since those will be naturally resolved by the market place.

In general, a level of FPs has to be tolerated, even by the most aggressive activists if the company can reduce a lot of fraud, but FPs are still serious.

Reply on news.ycombinator.com