Online reviews and ratings are important for potential customers when deciding whether to purchase a product or service. However, reading and synthesizing the massive amount of review data, which is often unstructured, is a huge challenge. In this study, we investigate the use of machine learning models to predict rating polarity (positive, neutral or negative) through automatic classification of review texts. We apply various single and ensemble classifiers to identify rating polarity of reviews from the 2017 Yelp dataset. Experimental results show that the linear kernel Support Vector Machine, Logistic Regression and Multilayer Perceptron are among the three best single classifiers in terms of accuracy, precision, recall and F-measure. Their performances can be further improved when used as base classifiers for ensemble models.