A Trial for Detection of Change in AS Operation Policies by Use of Machine Learning

The full route obtained by Border Gateway Protocol operation does not only indicate routes to all Internet hosts, but also has global information of the Internet structure. These are stored as information about AS operation policy in a condensed form in various attributes that the origin ISP or transit ISP gives to the prefix. This information is included statistically as well as analytically. Our preliminary research has shown that the network topology can be reproduced from the learning data on full root information obtained by a small-scale network model. Now, for the 100AS model created according to the Internet model configured using CAIDA data to collect and analyze the actual full route, the policy of AS including the topology can be restored. We also discuss whether such information can be reproduced by analyzing actual Internet data.