Integrating structural features for link prediction in directed social network using supervised machine learning with SHapley Additive exPlanations

Link prediction is a crucial task in social networks where the objective is to predict the missing links and upcoming links among the social entities. This paper considers the problem of link prediction as a classification problem. The network features are embedded into a matrix or dataset where the columns are network measures and rows are the edges represented by edge source and destination. Here, we consider the combination of local, quasi-local, and global measures as features. The dataset is divided into training and testing sets. Four supervised learning algorithms: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) are used for this task. The results show that the accuracy of XGBoost outperforms the rest three algorithms. The Shortest_Path is found to be the most contributing feature of this prediction by SHapley Additive exPlanations (SHAP). By this analysis, it is found that Shortest_Path has an interaction effect with Destination core number (Core_Num_Dst).