C-Command Software Forum

Accuracy 91.1% Correct - Is this good, bad or about right?

Screenshot 2017-10-04 17.26.55.png

Is this figure okay? I’m just curious.

Kind Regards


Oh dear, I can see that the accuracy is much higher than that. Please excuse the post. I’m just tired and probably can’t read the numbers correctly.

My accuracy now has fallen to 98.8%. Is there anything I should look at as to why this is?

Filtered Mail
4,657 Good Messages
203 Spam Messages (4%)
1 Spam Messages Per Day

SpamSieve Accuracy
2 False Positives
54 False Negatives (96%)
98.8% Correct

Corpus
2,082 Good Messages
205 Spam Messages (9%)
184,969 Total Words

Rules
245 Blocklist Rules
2,299 Whitelist Rules

Showing Statistics Since
13/07/2017, 02:16

Did you purposely train it with lots of extra good messages? That’s not recommended.

You could also look at the log to see why the spams are getting through or send in a diagnostic report.

I did train a lot of good messages initially. Would it be best to reset SpamSieve? If so, what shall I do?

Kind Regards

It looks like the extra training happened after you posted the initial screenshot.

I suggest that you send in a diagnostic report so that I can see why the spams are getting through.

Do you have a good supply of spam and good messages on-hand so that it would be easy to re-train SpamSieve if you reset it?

No, Michael, I don’t. I decided to do a completely fresh install as I don’t believe I followed the initial training correctly. The only issue I have since doing a clean reinstall is that the old statistics pane is still showing. Shouldn’t this be completely blank? For instance, it is still showing the following.

Filtered Mail
4,661 Good Messages
204 Spam Messages (4%)
1 Spam Messages Per Day

SpamSieve Accuracy
2 False Positives
54 False Negatives (96%)
98.8% Correct

Corpus
0 Good Messages
0 Spam Messages
0 Total Words

Rules
245 Blocklist Rules
2,371 Whitelist Rules

Showing Statistics Since
13/07/2017, 02:16

Shouldn’t the date and percentages have changed?

I did very carefully follow the uninstallation, removing the correct files etc. https://c-command.com/spamsieve/manual#uninstalling-spamsieve

I installed again and the statistics are as they should be. I will retain (very carefully) and get back to you in a week or so.
Thanks as always for your kind and quick responses on these forums. It’s much appreciated.

Yes, if you delete the History.db and Rules files as described in Removing SpamSieve’s Data Files.

Great!

Hi Michael,

I began again on the 8th November. Rather than bulk train existing messages, I simply trained them as they came in. Once I hit 100 good messages, I only trained spam from then on. In reality, I have trained all my good messages because anything that isn’t in my contact list is most probably spam. Slowly my accuracy has climbed as I approach the 65% spam and 35% good ratio.

Once I hit the 65% spam, I assume my accuracy will be at 100% (or as close as possible). But what happens after that? If I continue to train only spam, the ratio is going to be wrong and the accuracy will drop. Is that right?

Filtered Mail
404 Good Messages
117 Spam Messages (22%)
3 Spam Messages Per Day

SpamSieve Accuracy
0 False Positives
34 False Negatives
93.5% Correct

Corpus
100 Good Messages
120 Spam Messages (55%)
21,955 Total Words

Rules
82 Blocklist Rules
344 Whitelist Rules

Showing Statistics Since
08/11/2017, 20:58

My advice is to either do the initial bulk training or to use the Auto-train with incoming mail feature and only manually train the mistakes.

The accuracy depends on the total number of messages as well as the specific messages chosen, not just the ratio.

I recommend only training the mistakes, and SpamSieve will manage the ratio for you.

Since my last entry, I have been following your advice and only training the mistakes (It isn’t wrong very often). The accuracy is slowly climbing now.

Filtered Mail
553 Good Messages
182 Spam Messages (25%)
3 Spam Messages Per Day

SpamSieve Accuracy
1 False Positives
36 False Negatives (97%)
95.0% Correct

Corpus
124 Good Messages
186 Spam Messages (60%)
26,441 Total Words

Rules
90 Blocklist Rules
434 Whitelist Rules

Showing Statistics Since
08/11/2017, 20:58

Just very occasionally marking a new message as spam. The accuracy is still climbing. I hope it then stays at 100%

Filtered Mail
852 Good Messages
252 Spam Messages (23%)
3 Spam Messages Per Day

SpamSieve Accuracy
1 False Positives
37 False Negatives (97%)
96.6% Correct

Corpus
171 Good Messages
258 Spam Messages (60%)
32,585 Total Words

Rules
96 Blocklist Rules
500 Whitelist Rules

Showing Statistics Since
08/11/2017, 20:58