Personal spam control seems to go two ways. One using MAPS RBL which is basically a live list with a log of IP addresses being used for spam control. The other solution is using these Bayesian statistical spam control solutions which allow you to train your spam application over time.
Plenty of products are available for the latter.
I am confused as to which method to adopt. I could do a trial and error but I rather not. Also product recommendations would help. And finally, the product should have a low level of false positives.