We mostly rely on the spamd default settings, those are not easy to get right on a generic level. On my personal Cloudron I have tweaked those with great success, but mileage may vary.
In my case I have raised the score for some bayes rules to give the learning more relevance. You can try this on your instance by putting the following content:
score BAYES_999 2.0
score BAYES_99 4.5
into the file /app/data/spamd/custom.cf in your mail containers data folder, via SSH onto your host server the shell session is:
docker exec -ti mail /bin/bash
root@e05ce221283c:/app/code/haraka# cat /app/data/spamd/custom.cf
score BAYES_999 2.0
score BAYES_99 4.5
root@e05ce221283c:/app/code/haraka#
Then restart the mail container via the services view in your Cloudron dashboard.
As said though this may also result in mails landing in spam which are legit for your cases.
