Boosting Accuracy of Classical Machine Learning Antispam Classifiers in Real Scenarios by Applying Rough Set Theory

Table 2

Commonly used publicly available spam corpora.

Corpus	% legitimate	% spam	Number of messages
LingSpam^1	83.3	16.6	2893
PU1^1	56.2	43.8	1099
PU2^1	80.0	20.0	721
PU3^1	51.0	49.0	4139
PUA^1	50.0	50.0	1142
Spambase^2	39.4	60.6	4601
2005 TRECSpam^3	43.0	57.0	92189
2006 TRECSpam^3	35.0	65.0	37822
2007 TRECSpam^3	33.5	66.5	75419
SpamAssassin^4	74.5	25.5	9332

Available at https://labs-repos.iit.demokritos.gr/skel/i-config/downloads/.
²Available at http://ftp.ics.uci.edu/pub/machine-learning-databases/spambase/.
³Available at http://trec.nist.gov/data/spam.html.
⁴Available at https://spamassassin.apache.org/publiccorpus/.