Popfile Update #5

I’ve now been using Popfile’s Bayesian spam filter for more than a month. Accuracy has leveled off at about 95%, training on over 10,000 messages. The biggest shift in the last couple weeks is that most of the errors are now false negatives (missed spam), instead of good email filtered out. I haven’t had a false positive in a few days, which makes me think Popfile might be a worthwhile long-term solution. Dealing with a few spams that make it through is much easier than having to scrutinize the Popfile list every day for false positives.

The one other negative of Popfile is that its proxy, along with Radio Userland, frequently slow my machine to a crawl. I’m hoping the new Centrino-based laptop I have on order will alleviate this.