Improving spam mail filtering using classification algorithms with discretization Filter

1 ME Computer Student, 2 Professor 1,2 Computer Department, Pimpri Chinchwad College of Engineering, Nigdi, Pune, Maharashtra, INDIA _________________________________________________________________________ Abstract: Email spam or junk e-mail is one of the major problems of the today's usage of Internet, which carries financial damage to organizations and annoying individual users. Among the approaches developed till date to stop spam, filtering is an important and popular technique. Common practices for mail filters include organizing incoming email and removal of spam mails and viruses. A rare use is to examine outgoing email at some companies to ensure that employees observe with appropriate rules. Users might also employ a mail filter to prioritize messages, and to sort them into folders based on subject matter or other criteria. For many years these elements have driven pattern recognition and machine learning communities to keep improving email filtering techniques. Mail filters can be connected by the user, either as separate packages, or as part of their email program as email client. In case of emails, users can make manual filters which automatically filter mail according to the criteria chosen by particular user. In this paper, we present a survey of the performance of six machine learning methods in spam filtering techniques. Experiments are carried out on different classification techniques and association techniques using Waikato Environment for Knowledge Analysis (WEKA). Different classifiers are applied on one benchmark dataset in to evaluate which classifier gives better result. The dataset is in Attribute Relation File Format (ARFF). 10 fold cross validation is used to provide well accuracy. Results of classification algorithms are compared on spambase UCI dataset and it is found that no single algorithm performs best for spam mail filtering. For the different dataset it is observed that performance varies with different data sets. Our results prove that the performance of classification improve with filters.