Integrating machine learning techniques for critical node identification in complex networks

Identifying the most prominent nodes in complex networks becomes more critical for applications such as information propagation, epidemic control, and network robustness. In network structure analysis, centrality measures typically use the network's intrinsic topological structure and fail to capture the nonlinear dependencies between nodes structural features and their spreading ability under dynamic transmission scenarios. To overcome these limitations, this study proposes a machine learning based approach for efficiently identifying the most prominent in transmission scenarios. A feature vector is constructed for each node by integrating infection rate a crucial factor in spreading dynamics and various topological features. Later, the true spreading ability of each node, determined from propagation simulations using SIR and IC model is used for labelling. Several machine learning techniques, including Support Vector Machines, KNN, Random Forests, are evaluated as standalone classifiers. In addition, to better capture complex relationships between node features and spreading ability, we develop a hybrid clustering-classification approach that unifies K-means clustering with SVM (SVM+K-means), in which K-means is used to cluster nodes based on their feature similarity and SVM performs the final classification. Experimental results shows on multiple real world networks that our proposed frame work outperforms traditional centrality measures in accurately identifying influential nodes. The findings show that combining machine learning approaches with structural and dynamic network properties results in an effective and scalable strategy for identifying the most essential nodes. Our proposed frame work improves the accuracy of identifying influential nodes with minimum of 15% and maximum of 45% when compared with traditional centrality measures.