A System to Detect Forged-Origin BGP Hijacks

In this page, we show the daily accuracy of DFOH. More precisely, every day, we use our balanced sampling scheme to sample 1000 existing links and 1000 nonexistent links and run DFOH on these cases to evaluate its accuracy in various attack scenarios.

We show the True Positive Rate (TPR) and the False Positive Rate (FPR) in every attack scenario on the grids below. For every attack scenario, we also write below the TPR and FPR the number of links on which we tested DFOH.

Below, you can indicate on which day you want to see the accuracy of DFOH. Additionally, you can specificy which suspicious level to use (1, 5 or 10) and whether one features should be omitted, which is useful to check how and in which scenario the omitted feature is useful (or not).

For every new edge, this parameter indicates the minimum proportion of AS paths that DFOH needs to classify as suspicious in order to classify the new edge as suspicious (default is 50%).

Below you can indicate whether you want to keep all features or omit one of them

True Positive Rate

False Positive Rate

Information about the clusters