Analyzing the Efficacy of Machine Learning Models Towards Forecasting the Influence Spread of Nodes in Complex Networks

The identification of influential nodes in social networks is a favoured research area, owing to its use in myriad domains, especially with regards to viral marketing. Centrality measures are the most widely used to assess the influencing capability of nodes, though with the increase in network size the process becomes resource-intensive. In recent times, machine learning (ML) models are also being used to predict the influence spread of nodes, thereby aiding towards the identification of influential nodes. This paper evaluates the efficacy of diverse ML regressor models, employed to forecast the influence potential of the nodes, by taking various features as input vectors in a propagating scenario, under two real-world complex networks. This framework initially generates the feature vectors corresponding to each node using the node's structural and temporal aspects, and infection rates. It employs regression-based ML algorithms to learn the connection between the expected spreading ability of each node and its structural characteristics. Light Gradient Boosting Machine Regressor outperformed the other models with an average mean absolute and mean square error of 0.1232 and 0.1079, respectively.