Bayesian Root Cause Analysis by Separable Likelihoods
Root Cause Analysis for anomalies is challenging because of the trade-off between the accuracy and its explanatory friendliness, required for industrial applications. In this paper we propose a framework for simple and friendly RCA within the Bayesian regime under certain restrictions (namely that Hessian at the mode is diagonal, in this work referred to as separability) imposed on the predictive posterior. Within this framework anomalies can be decomposed into independent dimensions which greatly simplifies readability and interpretability.
We show that the separability assumption is satisfied for important base models, including Multinomial, Dirichlet-Multinomial and Naive Bayes. To demonstrate the usefulness of the framework, we embed it into the Bayesian Net and validate on web server error logs (real world data set).
KeywordsBayesian modeling Anomaly detection Root Cause Analysis
- [Deh17]Dehaene, G.P.: Computing the quality of the Laplace approximation, November 2017. http://adsabs.harvard.edu/abs/2017arXiv171108911D
- [KF09]Koller, D., Friedman, N.: Probabilistic Graphical Models: Principles and Techniques - Adaptive Computation and Machine Learning. The MIT Press, Cambridge (2009)Google Scholar
- [Pad14]Padmanaban, H.: Comparative analysis of naive Bayes and tree augmented naive Bayes models (2014). http://scholarworks.sjsu.edu/etd_projects/356
- [Shl14]Shlens, J.: Notes on Kullback-Leibler divergence and likelihood. http://arxiv.org/abs/1404.2000 (2014)
- [SMRE17]Solé, M., Muntés-Mulero, V., Rana, A.I., Estrada, G.: Survey on models and techniques for root-cause analysis. http://arxiv.org/abs/1701.08546 (2017)
- [Tu15]Stephen, T.: The Dirichlet-multinomial and Dirichlet-categorical models for Bayesian inference (2015). https://people.eecs.berkeley.edu/~stephentu/writeups/dirichlet-conjugate-prior.pdf