Predicting cascading failure with machine learning methods in the interdependent networks

Today one of the main challenges in interdependent networks is cascading failure. In infrastructure networks due to the nature of these networks, they usually depend on other infrastructure networks. Researches in this field are mostly focused on methods for evaluating networks and finding vulnerabilities according to the structure of the network. In this research, we pay attention to this problem from another point of view that is to predict before the failure occurs and then use algorithms to retrieve the network as soon as possible. The input data is randomly generated by the standard cascading failure simulation algorithm, the CFS algorithm. We teach our regression model with numbers of CFS reports that show how much of the network disrupts, when attack accrued on each node of the network. In this way we can predict the problem of cascading failure before it occurs in the network. This regression model employs node centralities in the network. This method is able to predict up to 80% of attacks that lead to failures above 10% and to recover the network quickly using the methods available for recovery.