Online Social Networks enables various users to connect and share their messages publicly and privately. On one hand it provides advantages to the users to connect and share but on the other hand it provides disadvantage of being attacks or post messages which contains negative or abuse words. Hence OSN provides various filtering rules for security against these wall messages. Although there are various filtering rules and classifiers implemented for the filtering of these users wall messages in popular OSN such as Twitter and Facebook. But in the proposed methodology not only filtering of these wall messages is done but the categorization of normal or negative messages are identified and hence on the basis users can be blacklisted. The proposed methodology is compared with FCM and SVM for clustering and classification of messages. This approach efficiently categorizes the messages but restricts for generating filtering rules and blacklist management. Thus the approach with FCM and J48 first initializes clustering using FCM followed by generation of rules using J48 based decision tree. Hence on the basis of the rules generated message are classified and message which doesn't contain attacks is then filtered on the basis of dictionary which contains a list of abuse words. The methodology is implemented by applying FCM and SVM and a comparison is done with FCM and J48 for the performance on the basis of accuracy to detect abnormal messages.