[Spambayes] Is Equal Ham & Spam really the best?

[skip at pobox.com]

    Dave> OK, I have some hams that scored as high as 0.85

I will assert without any further evidence that you have one or more
classification mistakes in your database.  Can you post the evidence header
from that message?

    Dave> Is everything between the ham_cutoff and spam_cutoff classified as
    Dave> unsure?  Moving my spam_cutoff to 0.86 probably would make only a
    Dave> small dent in my results.

If everything is properly trained SpamBayes should produce a distribution of
scores with a bimodal distribution, hams near 0.0 and spams near 1.0.  A ham
scoring above 0.5 tells me that you have classified a number of hams as
spam.

Skip