To design, develop and evaluate a system that predicts probable, but currently undiscovered, links within a social network graph via the use of machine learning and social network analysis techniques.
Link prediction algorithms attempt to predict the state of a graph at a given time t through various techniques. This report looks at several methods of link prediction on social network graphs including the use of metrics and machine learning. In this paper the link prediction problem has been posed as a classification model with Bagging Classifier filled with Support Vector machines being used as one of the classifiers. Evaluation of the project involved evaluating whether the classifier could recognise existing links and therefore whether it would be able to predict currently undiscovered links. The results demonstrate links can be predicted based solely on topological features.
Machine learning and data analytics can produce interesting and useful results however sometimes the data used is not complete. The use of link prediction helps to discover links that are currently not confirmed. The uses of link prediction could include helping people connect with eachother on social media, connecting researchers to collaborate with eachother based on similar previous connections and also in crime by helping to better map out a criminal enterprise.