Decision Theory For Large Scale Outlier Detection Using Aleatoric Uncertainty: With a Note on Bayesian FDR

TL;DR

This paper introduces a decision theoretic framework for large-scale outlier detection that leverages aleatoric uncertainty within Bayesian models, incorporating Bayesian FDR control to improve detection in nonstationary data.

Contribution

It presents a novel decision theoretic approach for outlier detection using aleatoric uncertainty and extends Bayesian FDR control for multiplicity adjustment in this context.

Findings

Effective in nonstationary, temporally fluctuating outlier scenarios

Improves detection performance over fixed threshold methods

Validated on cybersecurity threat detection case study

Abstract

Aleatoric and Epistemic uncertainty have achieved recent attention in the literature as different sources from which uncertainty can emerge in stochastic modeling. Epistemic being intrinsic or model based notions of uncertainty, and aleatoric being the uncertainty inherent in the data. We propose a novel decision theoretic framework for outlier detection in the context of aleatoric uncertainty; in the context of Bayesian modeling. The model incorporates bayesian false discovery rate control for multiplicty adjustment, and a new generalization of Bayesian FDR is introduced. The model is applied to simulations based on temporally fluctuating outlier detection where fixing thresholds often results in poor performance due to nonstationarity, and a case study is outlined on on a novel cybersecurity detection. Cyberthreat signals are highly nonstationary; giving a credible stress test of the model.