Efficient link prediction model for real-world complex networks using matrix-forest metric with local similarity features

Link prediction in a complex network is a difficult and challenging issue to address. Link prediction tries to better predict relationships, interactions and friendships based on historical knowledge of the complex network graph. Many link prediction techniques exist, including the common neighbour, Adamic-Adar, Katz and Jaccard coefficient, which use node information, local and global routes, and previous knowledge of a complex network to predict the links. These methods are extensively used in various applications because of their interpretability and convenience of use, irrespective of the fact that the majority of these methods were designed for a specific field. This study offers a unique link prediction approach based on the matrix-forest metric and vertex local structural information in a real-world complex network. We empirically examined the proposed link prediction method over 13 real-world network datasets obtained from various sources. Extensive experiments were performed that demonstrated the superior efficacy of the proposed link prediction method compared to other methods and outperformed the existing state-of-the-art in terms of prediction accuracy.