Link Prediction Method Using Topological Features and Ensemble Model

Online social networking has progressively been the new interdisciplinary research area, especially for developing new strategies of investigating these informal networks containing billions of users. However, such networks might not represent real-world connections among people either due to imperfect procurement forms or not yet reflected on the online platform like friends in real-world might not connect with each other online. To predict these unknown connections in the online community is still an open-ended problem. In this paper, a novel link prediction method is proposed to find the missing connections in the social network graphs. The proposed method extracts topological features from the network graph which are used to train an ensemble learning model i.e., random forest classifier. The trained model is used to predict the missing connections. The experimental evaluation is conducted on two networking dataset namely; ‘Facebook networking dataset’ and the ‘Flickr following dataset’ publicly available on Stanford Network Analysis Project (SNAP) and Koblenz Network Collection (KONECT) respectively. The comparison is done with the prediction results on the same features by the state-of-the-art learning models namely; linear support vector machine (LSVM), K-Nearest Neighbours (KNN), AdaBoost, and Gradient Boost. The performance of the considered methods is defined in terms of accuracy, precision, recall, F1-measure, and AUC value. Additionally, the efficiency of the proposed method is validated against the existing link prediction method. The experimental results conclude that the proposed method is accurate than the compared methods in uncovering the hidden links of a social network.