A lethal online substance has turned into a noteworthy issue in this day and age because of an exponential increment in the utilization of the web by individuals of various societies and instructive foundation. Separating hate speech and offensive language is a key test in the programmed detection of dangerous content substance. In this paper, we propose a way to deal with naturally order tweets on Twitter into three classes: hateful, offensive and clean. Utilizing Twitter dataset, In this paper, we propose a way to deal with distinguish hate expressions on Twitter. Our methodology depends on unigrams and examples that are consequently collected from the preparation set. These examples and unigrams are later utilized, among others, as highlights to prepare a machine learning calculation. Our analyses on a test set made out of 2010 tweets demonstrate that our methodology achieves an exactness equivalent to 87.4% on identifying whether a tweet is offensive or not (twofold classification), and precision equivalent to 78.4% on distinguishing whether a tweet is hateful, offensive or clean (ternary classification).