A novel similarity-based parameterized method for link prediction

Any complex real-world system that changes over time can be represented as a network. We analyze these networks using network theory-based techniques to infer useful information from them. An important problem associated with complex systems is the link prediction problem. It aims to find the possibility of future or missing links in a network. Existing similarity-based link prediction methods consider one or two network features for link prediction and perform well on specific types of networks. This empirical work proposes a novel similarity-based parameterized algorithm for link prediction in complex networks. The proposed method uses three simple features and performs well on the various categories of networks. Using AUC (Area Under the Receiver Operating Characteristics Curve), accuracy, and f-measure as the performance metrics, we conduct an experimental evaluation of the proposed method against nine state-of-the-art methods and on five real-world datasets. We also perform a time comparison of the proposed method against others. It is more accurate and time-efficient compared to recent learning-based methods. The experimental results assert the enhanced performance of the proposed method.