We use groups (users) and get a lot of spam,
so for what it's worth (took a few months to
get over 99%, has been running for a year):
users:
TP True Positives: 41349
TN True Negatives: 149947
FP False Positives: 123
FN False Negatives: 865
SC Spam Corpusfed: 0
NC Nonspam Corpusfed: 0
TL Training Left: 0
SHR Spam Hit Rate 97.95%
HSR Ham Strike Rate: 0.08%
OCA Overall Accuracy: 99.49%
We're happy.
/L
Glynne Jones wrote:
> I'm seeing the same behaviour and I have a lot of good mail:
>
> TP True Positives: 394
> TN True Negatives: 3205
> FP False Positives: 3
> FN False Negatives: 276
> SC Spam Corpusfed: 520
> NC Nonspam Corpusfed: 4
> TL Training Left: 0
> SHR Spam Hit Rate 58.81%
> HSR Ham Strike Rate: 0.09%
> OCA Overall Accuracy: 92.81%
>
> Regards,
>
> Glynne
>
>
>
> On 6 Oct 2008, at 13:44, Kenneth Marshall wrote:
>
>> You probably need to train with more good mail. DSPAM knows what
>> bad mail contains, but not what is in good mail. The end result
>> is much, much less effective filtering.
>>
>> Ken
>>
>> On Mon, Oct 06, 2008 at 09:33:28AM +0530, Ranbir Sanasam wrote:
>>> Hi,
>>> I am facing a strange problem (effectiveness of DSPAM). I have few users
>>> whose overall accuracy is just about 70%.
>>> Following is the dspam_stat
>>> TP True Positives: 3625
>>> TN True Negatives: 3589
>>> FP False Positives: 109
>>> FN False Negatives: 1910
>>> SC Spam Corpusfed: 2203
>>> NC Nonspam Corpusfed: 0
>>> TL Training Left: 0
>>> SHR Spam Hit Rate 65.49%
>>> HSR Ham Strike Rate: 2.95%
>>> OCA Overall Accuracy: 78.13%
>>>
>>> This user had forwarded around 2000 spam mails to spam alias and
>>> around 100
>>> notspam alias for spam and not spam misclassification respectively.
>>> And he
>>> had been participating from the beginning of the DSPAM installation.
>>> I have
>>> set training mode TEFT, database MYSQL. we have waited observing
>>> till its
>>> training left equals to 0 so as to kick start other DSPAM features.
>>> Still,
>>> its accuracy remain around 60 to 70%.
>>> Please help me to find the solution. Any suggestions are welcome.
>>>
>>> Interestingly, for some users, accuracy is above 99%. But, for some
>>> participating users, accuracies are just about 70%.
>>> Please help me.
>>>
>>> Thanking you
>>> ranbir
>>>
>>>
>>>
>>
>>
>>
>>
>
>
>
>
>
>
!DSPAM:1011,48eb4251150921343813492!
