Description
When we study epidemic spread processes, one of the most important problems is to give an accurate estimate for the infection parameter. In the current work, our main goal is to quantitatively compare different statistical methods for this problem in the case when the underlying network is a two-layer random graph, where the first layer represents households, the second layer is for modelling weaker connections between the members of different households. In this setup, we analysed the behavior of estimators based on the maximum likelihood method, the xgboost method and neural networks. We studied the effect of the structure on the underlying random graph on the quality of the estimators, and also examined which statistics are useful for the estimations, for example, whether it is worth collecting data from households, and not just from individuals. We also studied similar questions for hypergraphs and opinion spread processes, where higher-order interactions play an important role. In the talk, we present the models that we used, the results of our computer simulations and the main conclusions for these statistical problems.
Joint work with Edit Bognár, Villő Csiszár, Damján Tárkányi and András Zempléni.