In recent years, hate speech moved specially due to their simple and anonymous distribution through social networks more and more into the focus of the society and is now constituting a non-negligible problem. Especially in a social medium like Twitter, the large amounts of posts can only be inadequately investigated for derogatory content or offensive language in order to react accordingly. Therefore an approach, based on supervised machine learning, is presented in this work, which identifies hate postings automatically. For this to be accomplished, features already proven in previous works for the recognition of offensive remarks, which consider characteristics of the used language and the Tweet content, were included. In addition, special attention was paid to features that can be gained by analyzing the network structure and the use of a dictionary customized for the hate identification. Finally, the model of a classifier is trained with the resulting features, which classifies a Tweet as neutral or hateful. In detail a Support Vector Machine-, Naive Bayes and Random Forest- classifier came to use. To evaluate the performance of the machine learning algorithmus different experiments were carried out, which should give information about how features and its combinations affect the accuracy of the classifications, as well the respective classifiers perform and how the parameters of these have to be adjusted to optimize the results further. Based on the calculated values the combination of feature set and classifier with its optimal parameter settings, from which one expects the best identification of hate postings, is presented as the final result of this work.