Millions of users worldwide participate in social networking sites. The interactions between users and those social sites, as Facebook and Twitter, have an enormous effect as well as sometimes unwanted impact on daily life. Spammers have become a target platform on the prominent social networking sites to scatter huge amounts of damaging and irrelevant information. Twitter has become one of the best-used platforms of all times also thus permits unreasoning spamming. Fake users send unwanted tweets to users to encourage services or sites that affect not just genuine users, but also disrupt the consumption of resources. The main aim of this study is to identify forged users who harm real users, threaten users’ identity and therefore their safety and privacy. In this paper, suggest to URL based detection plus machine learning techniques has been developed for the detection of malicious users, celebrities as well as malicious users, based on user characteristics. A crawler is developed for Twitter for the purpose of detecting malicious users, celebrities and non-malicious users and celebrities, and data from around 22 K users were collected from public information. For training and testing purposes, data from around 7,500 users were used in machine learning classifications to classify users. On basis of performance metrics like accuracy, recall, F-measurement also accuracy 5 classifiers were used and compared. With a precision of 99.8 percent, Random Forest is above all classifiers.