A novel and precise approach for similarity-based link prediction in diverse networks

In recent times, real-world systems have been rapidly growing in size resulting in increased complexity. Networks are an interpretation of the complex system which describes the structure of a complex system by understanding the relations between the elements. Link prediction plays an important role in network analysis through which we can observe the hidden or missing link between the nodes. In this paper, we have proposed an improved hybrid similarity-based link prediction approach. We use five different metrics to experimentally evaluate the performance of the proposed approach such as AUC, precision, recall, F1 score, and accuracy. We also compare the proposed approach with recent and existing link prediction approaches against six real-world datasets. We find that the proposed approach performs well for all the considered metrics as compared to other existing link prediction approaches. Additionally, we also compare the proposed approach against state-of-the-art link prediction approaches using classification-based machine learning algorithms such as logistic regression, random forest, k-nearest neighbors, and naïve Bayes. The results show that the proposed approach outdoes the other link prediction approaches in terms of AUC. The proposed method has potential practical use for network analysis in various networks such as information networks, technological networks, and social networks.