Information theory for data-driven risk analysis: The informational coefficient of correlation as a measure of dependency

Publikation: Bidrag til bog/antologi/rapport/proceedingKonferencebidrag i proceedingsForskningpeer review

In this paper uncertainty quantification and structural reliability analysis are reinterpreted through the information theory. At first, main concepts of information theory are presented, e.g. Entropy, KL Divergence, Mutual Information (MI); their relationships with the classical uncertainty quantification, like maximum likelihood estimation and copulas are discussed. Moreover, the Distributions with Independent Components (DIC), recently proposed by the authors for uncertainty quantification, are based on the mutual information. Herein, it is questioned the adoption of the linear coefficient of correlation to measure the dependence between random variables. An alternative measure, called informational coefficient of correlation and based on the mutual information is suggested. Some examples show the robustness of the proposed informational coefficient of correlation, which can be used also as a metric for global sensitivity analysis.

OriginalsprogEngelsk
TitelProceedings of the 29th European Safety and Reliability Conference, ESREL 2019
RedaktørerMichael Beer, Enrico Zio
Antal sider8
ForlagResearch Publishing Services
Udgivelsesår2020
Sider2113-2120
ISBN (Elektronisk)9789811127243
DOI
StatusUdgivet - 2020
Begivenhed29th European Safety and Reliability Conference, ESREL 2019 - Hannover, Tyskland
Varighed: 22 sep. 201926 sep. 2019

Konference

Konference29th European Safety and Reliability Conference, ESREL 2019
LandTyskland
ByHannover
Periode22/09/201926/09/2019
SponsorExida.com LLC, Grossraum-Verkehr Hannover GmbH, Safety Tools Development (SATODEV)

Se relationer på Aarhus Universitet Citationsformater

ID: 194905468