Comparison of Machine Learning Algorithms for English and Japanese SPAM Mail Discrimination

• 作 者: NatsukiFujimori;ShinichiYoshida;
• 机 构: Kochi University of Technology
• 出 处: 第三届先进计算智能与智能信息学国际研讨会(The Third International Workshop on Adv
• 发表日期: 2013年期

　　In order to reveal the ability of a classification algorithms to text data,Naive Bayes Classifier,Neural Network,Support Vector Machine(SVM),Bagging,AdaBoost,and Random Forest are applied to classify email written in both English and Japanese.Those algorithms are compared with each other from a viewpoint of classification precision for text data that is categorical and discrete data of word frequency appeared in the texts.For English e-mail classification,the dataset “Spambase” of UCI Machine Learning Repository is used.Total number of the data is 4601,and training data is from 500 to 4000,which are randomly selected,and the rest are the test data.For Japanese email classification,original corpus is created and used.Total number of the data is 1400 and training data are randomly selected to 1000.SPAM email in English discrimination is performed under the same conditions to compare the result of precision.As a result,for English SPAM distinction,all algorithms except Naive Bayes Classifier achieves the precision exceeding 90%.Moreover,for Japanese SPAM,Naive Bayes Classifier became a distinction rate of 42.8%,and 77-79%obtained by other algorithms.