Finding future associations in complex networks using multiple network features

Finding future or missing associations is an essential problem in complex systems because of its numerous application areas. These applications use link prediction methods to provide services to their customers. Our work proposes a novel similarity-based approach for finding probable associations that uses the concepts of closeness and the degree of common neighbors. Available similarity-based approaches for finding probable associations in complex systems perform well on specific categories of networks. They utilize a limited number of features. Our designed method performs well on different categories of networks because it uses three simple features. We compare our approach against eight popular and recent link prediction approaches on six real networks of varied sizes. We conducted an experimental evaluation of all the approaches using AUC (Area Under the ROC Curve), accuracy, precision, recall, and F1-score as the performance measures. The experimental evaluation shows the effectiveness of our designed method, as it outperforms other existing link prediction techniques. It remarkably improves prediction capability.