Simple Root Cause Analysis by Separable Likelihoods

  • 2018-08-13 15:54:50
  • Maciej Skorski
  • 2

Abstract

Root Cause Analysis for Anomalies is challenging because of the trade-offbetween the accuracy and its explanatory friendliness, required for industrialapplications. In this paper we propose a framework for simple and friendly RCAwithin the Bayesian regime under certain restrictions (that Hessian at the modeis diagonal, here referred to as \emph{separability}) imposed on the predictiveposterior. We show that this assumption is satisfied for important base models,including Multinomal, Dirichlet-Multinomial and Naive Bayes. To demonstrate theusefulness of the framework, we embed it into the Bayesian Net and validate onweb server error logs (real world data set).

 

Quick Read (beta)

loading the full paper ...