The spread of Internet and social media led to be sentiment analysis an open research area. Social media is used so the people can be state their opinions and attitudes on blogs, Tweets, and forums. Sentiment analysis deals with identifying and extracting people's opinions and attitudes from texts on the internet. The classification of the text which is based upon sentiment is differ from topical text classification because it has recognition based on an opinion on a topic. This research studying the ability to apply TF-IDF feature selection approach for sentiment analysis and examines the performance for classification by 4 machine learning methods (naïve Bayes, KNN, J48, and logistic regression) with regard to recall, precision and F1-measure. This research included a comparison between the selected ML methods. The results show the naïve Bayes over performed on other classification methods with precision about 94.0%.