Incremental Naïve Bayesian Spam Mail Filtering and Variant Incremental Training

This paper proposes an incremental spam mail filtering using Naive Bayesian classification which gives simplicity and adaptability. To keep the training set to a limited size and small, the sliding window is applied and the training set is updated when new emails are received. In effect, features in the training set are incrementally updated, and the model would be adaptive to a new spam pattern. In addition, we present three incremental training schemes: a window containing only the most recent emails, a window containing the previous batch of emails, and a window containing all already seen emails. The proposed model is evaluated using two spam corpora: Trec05p-1 and Trec06p. In our experiments, the window size is varied, the processing time per message, and the ham and spam misclassification rates are measured. The results show that the third incremental training scheme gives the best outcomes, and the window size significantly affects the misclassification rates and the processing time.