Today, machine learning approaches are used to make computers act more accurately for various purposes. In this manner, one area in which the machine learning approaches are used is the detection of phishing web sites. Phishing is an online threat, which depends on creating a fake web site that mimics a trustworthy web site to steal important personal information. It is important to predict whether a website is a phishing website in order to avoid this danger before it happens. In this study, AdaBoost, multilayer perceptron, support vector machine, decision tree, k-nearest neighbors, Naïve Bayes and random forest machine learning techniques are compared to predict the purpose of a website. This comparison is performed by experimenting over a dataset containing 1353 instances with 9 different features. The experimental evaluation is performed in two different settings. The first setting based on splitting the data into training and test sets. In this setting the evaluation results show that the random forest algorithm, which is an ensemble learning approach based on decision trees, outperforms other compared approaches. On the other hand, in the second setting based on cross validation, multilayer perceptron shows a better performance.