Critical Episode Mining in Intrusion Detection Alerts

One of the most important steps in attack detection using Intrusion Detection Systems (IDSs) is dealing with huge number of alerts that can be either critical single alerts and multi-step attack scenarios or false alerts and non-critical ones. In this paper we try to address the problem of managing alerts via a multi-layer alert correlation and Itering that can identify critical alerts after each step of correlation and Itering. After applying the approach on LL DDoS 1.0 data set, we achieved very good results in terms of critical alert detection rates, running time of approach and its memory usage. Our method could extract all of critical and multi-step attacks in LL DDoS 1.0 data set while we had almost 90% reduction in number of alerts.