This project is to explore how private information might be inferred by using public friendship data on the Facebook social network with the hope that the result of the project could be used to help develop relevant algorithms regarding to privacy protection on social media platforms. First of all, some existing social network analysis methods, popular data analysis frameworks and social media data extraction approaches are reviewed. Based on an existing data set, nine prediction methods are applied in order to predict the private gender information. According to the result of the experiment in the project, the accuracy of prediction is above 70 percent on average. Among all the methods, multi-layer artificial neural network models perform best, which give a prediction accuracy of 79 percent. When users' major information is considered, the accuracy of prediction could reach up to 81 percent. In addition, how different parameter choices of prediction models might affect the result is discussed. Finally, based on limitations that have been found, relevant future work is suggested.