Increase in number of spam incidents is causing a very serious threat to Social Networking World which has in turn become an important means of interaction and communication between public users. It is not only dangerous to the public users, but it also covers much of the bandwidth of the Internet traffic. Most of current spam filters in use are based on the subject content of email, Facebook, twitter. Social Networking Services also provide great possibilities to take advantage of user identification and other social graph-dependent features to improve classification. In this paper, the proposed System uses machine learning [3] approach for spam detection based on features extracted from social networks constructed from social networking site message metadata and logs. Flags and scores are assigned to senders based on their possibility of being a legitimate sender or spammer. Moreover, proposed System also explores various spam filtering techniques and possibilities. Social networking sites are vulnerable to mass spam incidence as well as users data theft such as credit card details, user activities and users taste for criminal purposes . Email subject headers are used to check spam email, spam on Social networking Sites is often accompanied by a wealth of data on the sender, metadata can be used to build more accurate detection mechanisms. System uses these terminologies to choose features that best differentiate spammers from legitimate users. On basis of this technique system flag user system or message as spam and legitimate messages.

SNS (Social Network Services) [2] OSN (Online Social Network) [2] tf-idf (Term Frequency-Inverse Document Frequency).