An improved Bayesian filtering technique for spam recognition

Abstract

In this paper, we presented models and software for spam recognition using an improved Bayesian filtering technique. Based on a corpus from Androutsopoulos et al, our Spam Recognition framework outperforms other state-of-the-art learning methods based on Bayesian algorithm in terms of spam detection capability. Our software has proved an accuracy of 99.9% of good classification. The 0.1% of other messages have been classify as “may be spam” due to their vagueness signature. Brief, in the case of extremely high misclassification cost, our model still remains stable accuracy with low computation cost, while other methods’ performance deteriorates significantly as the cost factor increases. Keywords: Spam filter; filter technique; Bayesian algorithm; Bayes technique; Naive Bayes; Spamming techniques; Spamming preventive techniques.