In this paper, different machine learning algorithms, ensemble algorithms,and deep learning algorithms are applied to Arabic tweets to detect whether ithuman-generated or not. The tweets are used twice as preprocessed and nonpreprocessed to measure the effectiveness of Arabic preprocessing in theclassification process. The data is also tokenized with various methods like unigram,trigram, and Term Frequency–Inverse Document Frequency. The experimentsshow that the support vector machine with the non-preprocessed tweets andunigram tokenization has the best performance of 83.11% and a precision of 0.9516while it predicts the spam or not in a relatively small time.