looks very much like pruning in decision trees :
1) http://www.cs.cmu.edu/~tom/mlbook.html chapter III
2) Quinlan, J. R. C4.5: Programs for Machine Learning. Morgan Kaufmann
Publishers, 1993.

Sorry, a little bit too long to explain in a mail.

The idea is :
knowing the number of errors (E) among N tries, find the error proba such that
seeing E errors among N is not rare.
binomial law -> approximated with normal law -> confidence intervals

Approximating binomial with normal law can be done when the sample size is circa
35

Hope this helps

Post a reply to this message