I was reading about statistical filters, and I came across a sentence that I didn't understand:

You might be thinking that there's not much difference between 95 percent and 99.5 percent. But you would be wrong. A filter with a 99.5 percent accuracy rate is not merely 4.5 percent more effective than one that is 95 percent accurate, but 900 percent more effective!

Can someone explain how a filter that is 99.5 percent accurate is 900 percent more effective than 95 percent?

  • 1
    $\begingroup$ This is from a book called Ending Spam: Bayesian Content Filtering and The Art of Statistical Language Classification $\endgroup$ – S. Sharma Mar 27 at 14:43

Let's consider a system with a $95~\%$ certainty. It means that for every $10~000$ inputs (or cases, whichever you want), there would be about $(1-0.95)\times 10~000\approx 500$ errors.

Then compare it to a system with $99.5~\%$ certainty. Now the number is $(1-0.995)\times 10~000\approx 50$.

The first system has $\frac{500-50}{50}\approx 900~\%$ more errors. That's what the text is trying to communicate.

PS: I put the "approximate" signs to indicate the expected number of errors.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.