Artificial immune system with spam filter

Since today, e-mails are fast and cheap enough, they engages the attentions of individuals and companies who want to advertise and make promotions. Person or companies send millions of unnecessary and unsolicited e-mails with the same content to unidentified e-mail accounts in order to reach their aims. Many people are subjected to this type of e-mails every day. These e-mails occupy internet network bandwidth too much along with having disturbing contents, loading a huge cost to the companies providing e-mail services and wasting the internet user?s times.This thesis aims to make contribution on the stable struggle about filtering unsolicited e-mails. In this work, it is aimed to detect unsolicited e-mails written in Turkish language. Bayes classifier and different algorithms of artificial immune system (AIS) are utilized and best AIS algorithms are determined for this study. In this thesis, AIRS1, AIRS2, AIRS2PARALLEL, CLONALG and CSCA algorithms of AIS, have been investigated.Turkish spam and Non-spam e-mails are collected. Initially these e-mails are splitted into two groups. Fisher dicriminant analysis and Eulidean distance methods are utilized for feature extraction. E-mails of each group are processed in order to build dataset from the feature vector. Thus, four different datasets are obtained from two types of feature extraction on the original two datasets. Classification algorithms are performed on these four datasets and then mean of the results are taken. It is observed that, classification rates of the Fisher based features outperform to Euclidean based features. CSCA algorithm which is a significant type of the AIS algorithms is determined as a best spam e-mail classifier.Key Words: Artificial Immune System, Unsolicited E-Mail, Dataset Generation, Fisher Discriminant Analysis, CSCA, AIRS