Link prediction for weighted collaboration networks based on nodal attributes

Abstract Link prediction in collaboration networks plays a vital role in understanding and fostering scientific collaborations. In this paper, we constructed a weighted collaboration network based on co-author information from papers published in 20 statistical journals between 2003 and 2024. By analyzing the largest connected component of the network, we evaluated the performance of various similarity indices in both unweighted and weighted settings, demonstrating the benefits of incorporating edge weights. Additionally, we proposed a novel prediction factor, the Combined Effect Score (CES), which integrates nodal attributes such as authors’ affiliations and research interests. To integrate all predictors, various machine learning methods such as Logistic Regression, XGBoost, and Random Forest have been applied. After integrating the similarity indices with the CES index, we observed substantial improvements in the Area Under the Receiver Operating Characteristic Curve (AUC) values across all classifiers, demonstrating that nodal attributes can significantly enhance prediction accuracy. This research not only advances link prediction methodologies but also provides actionable recommendations for fostering new academic collaborations, thereby facilitating the formation of impactful research partnerships.