Feature based link prediction

Under the different searches performed to analyzing social networks, much attention has been devoted to the problem of predicting links. It is a key technique in many applications such as recommendation systems which provide suggestions of potential links between nodes. Traditional link prediction methods use a single proximity metric. In this paper, we study link prediction as a supervised learning task where we try to combine multiple features as input data for classification. To improve the accuracy of prediction, we have been applying a select attributes algorithm. Experiments have been performed on two co-authorship data sets. Results demonstrate that Random Forest, k-nearest neighbors and Principal Component Analysis yield the best performances.