Using machine learning to identify epidemic threshold in complex networks

Abstract

Machine learning is a powerful tool for identifying the phase of matter. Usually when the phase information is fully marked, the direct application of supervised learning can successfully detect phase transitions, while the unsupervised learning method does not require any prior knowledge to distinguish phases of matter, and even discover new phases of matter. Here, we have developed a machine learning framework containing unsupervised learning ideas to identify phase transitions in the dynamics of epidemic spreading in complex networks. The framework trains the neural network so that the configuration information of the epidemic spreading dynamics can describe the effective spread rate, and the accuracy of the effective spreading rate predicted by the neural network can be used as an indicator of phase transition. Tests on a large number of synthetic networks and real networks have proved that the framework has low computational cost, high efficiency, and is suitable for complex networks of any size and topology.