Research on Spam Filtering Technology Based on New Mutual Information Feature Selection Algorithm

Abstract

Aiming at the deficiency of traditional mutual information algorithm in feature selection, this paper proposes a weighted naive Bayesian algorithm based on improved mutual information, called imi-wnb algorithm. In the feature selection stage, the word frequency factor and the difference factor between classes are introduced to improve the traditional mutual information algorithm to achieve feature dimension reduction. In the process of classification, the value of IMI is introduced to weight the attributes of naive Bayes algorithm, which partly eliminates the influence of conditional independence assumption of naive Bayes algorithm on classification, and improves the efficiency and stability of spam classification.