Using a machine learning tool in diagnosis of network overload

Abstract

Diagnosis of network anomalies is an important component of network management. A way to obtain a set of diagnostic rules by using a machine learning (ML) tool is described. In particular an overload situation is analyzed. Expert knowledge and a series of simulated network situations are employed together as inputs to a machine learning system. The ML tool helps the traffic engineer in proposing new rules and refining preexisting ones. The authors present a possible sequence of experiments that has led to a set of rules for recognizing a specific overload situation.<>