With the increasing volume of data on the web, the user needs a fast, accurate and directly access to information and quick responses to his/her questions. This led to motivating the authors of this paper to adopt the idea of Question Answering Systems; these systems depend widely on the language. There are many research studies have been conducted focusing on the English and Latin languages. However, number of studies related to the Arabic language is very limited and it still needs to more enhancements. Question Answering Systems consist of three main modules (Question Pre-processing, Document Processing and Answer Processing). Question preprocessing form is considered as an important phase on the overall Question Answering Systems because it plays a main role in identifying the question class which effects on the detection of a candidate answer. Within this research study, the focus was on identifying the Arabic question class based on an Arabic taxonomy and building a model for classifying the Arabic question using machine learning algorithms such as Support Vector Machine, Naive Bayes, and Logistic Regression., with an attempt to build a Named Entity Recognition model for Arabic language. Although, the results were not as good as expected, however there is a chance to improve the results in future. The outcomes of the experiment were as follows: logistic regression achieved the highest accuracy 82%, linear SVM 81% and Multinomial 79%.