Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

A Bayesian Local Causal Discovery Framework

Mani, Subramani (2006) A Bayesian Local Causal Discovery Framework. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Primary Text

Download (943kB) | Preview


This work introduces the Bayesian local causal discovery framework, a method for discovering unconfounded causal relationships from observational data. It addresses the hypothesis that causal discovery using local search methods will outperform causal discovery algorithms that employ global search in the context of large datasets and limited computational resources.Several Bayesian local causal discovery (BLCD) algorithms are described and results presented comparing them with two well-known global causal discovery algorithms PC and FCI, and a global Bayesian network learning algorithm, the optimal reinsertion (OR) algorithm which was post-processed to identify relationships that under assumptions are causal.Methodologically, this research formalizes the task ofcausal discovery from observational data using a Bayesianapproach and local search. It specifically investigates theso called Y structure in causal discovery andclassifies the various types of Y structurespresent in the data generating networks. Itidentifies the Y structures in the Alarm,Hailfinder, Barley, Pathfinder and Munin networks andcategorizes them. A proof of the convergence of the BLCDalgorithm based on the identification of Y structures, isalso provided. Principled methods of combiningglobal and local causal discovery algorithms to improve uponthe performance of the individual algorithms are discussed. In particular,a post-processing method for identifying plausible causal relationships from the output of global Bayesiannetwork learning algorithms is described, therebyextending them to be causal discovery algorithms.In an experimental evaluation, simulated data fromsynthetic causal Bayesian networks representing fivedifferent domains, as well as a real-world medical dataset, were used. Causal discovery performance was measured using precision and recall.Sometimes the local methods performed better than the global methods,and sometimes they did not (both in terms of precision/recalland in terms of computation time).When all the datasets were considered in aggregate,the local methods (BLCD and BLCDpk) had higher precision.The general performance of the BLCD class of algorithmswas comparable to the global search algorithms, implying that the localsearch algorithms will have good performance onvery large datasets when the global methods fail to scaleup. The limitations of this research and directions for future research are also discussed.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairCooper, Gregory Fgfc@cbmi.pitt.eduGFC
Committee MemberBuchanan, Bruce Gbuchanan@cs.pitt.eduBBUCHANA
Committee MemberWagner, Michael Mmmw@cbmi.pitt.eduMMW1
Committee MemberSpirtes,
Date: 30 March 2006
Date Type: Completion
Defense Date: 17 May 2005
Approval Date: 30 March 2006
Submission Date: 8 December 2005
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Institution: University of Pittsburgh
Schools and Programs: Dietrich School of Arts and Sciences > Intelligent Systems
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: BLCD; Causal Bayesian networks; Causality; Infant mortality; Local search; Markov blanket; Y structure; Causal discovery; Global search
Other ID:, etd-12082005-122145
Date Deposited: 10 Nov 2011 20:09
Last Modified: 15 Nov 2016 13:53


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item