Application research of machine learning algorithms for spam filtering

摘要

Bayesian filter technology based on machine learning theory is presently one of the most popular. Relatively speaking, the perceptron algorithm belongs to the nonparametric algorithm. Relative to the Bayesian algorithm commonly used, it does not need to make any assumptions on the statistical properties of all kinds of samples, which belongs to the deterministic method. So, the classification accuracy of the perceptron algorithm is higher than that of the Bayesian algorithm. Through the analysis of the perceptron algorithm and the averaged perceptron algorithm, two kinds of spam classifier based on them are trained, respectively. On the basis of analysis on the relationship between spam classifier parameters configuration and the performance of different algorithms, a better algorithm and parameter configuration used to design spam classifier is provided. Theoretical analysis and experimental results show that spam classifier based on the average perceptron algorithm has better performance in spam classification than spam classifier based on the perceptron algorithm. Therefore, it can be used to separate spam preferably.