11.07.2015 Views

Causality in Time Series - ClopiNet

Causality in Time Series - ClopiNet

Causality in Time Series - ClopiNet

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

JMLR: Workshop and Conference Proceed<strong>in</strong>gs 12:115–139, 2011<strong>Causality</strong> <strong>in</strong> <strong>Time</strong> <strong>Series</strong><strong>Time</strong> <strong>Series</strong> Analysis with the <strong>Causality</strong> WorkbenchIsabelle Guyonisabelle@clop<strong>in</strong>et.com<strong>ClopiNet</strong>, Berkeley, CaliforniaAlexander StatnikovAlexander.Statnikov@nyumc.orgNYU Langone Medical Center, New York cityConstant<strong>in</strong> AliferisConstant<strong>in</strong>.Aliferis@nyumc.orgNYU Center for Health Informatics and Bio<strong>in</strong>formatics, New York cityEditors: Flor<strong>in</strong> Popescu and Isabelle GuyonAbstractThe <strong>Causality</strong> Workbench project is an environment to test causal discovery algorithms.Via a web portal (http://clop<strong>in</strong>et.com/causality), it provides anumber of resources, <strong>in</strong>clud<strong>in</strong>g a repository of datasets, models, and software packages,and a virtual laboratory allow<strong>in</strong>g users to benchmark causal discovery algorithmsby perform<strong>in</strong>g virtual experiments to study artificial causal systems. We regularlyorganize competitions. In this paper, we describe what the platform offers forthe analysis of causality <strong>in</strong> time series analysis.Keywords: <strong>Causality</strong>, Benchmark, Challenge, Competition, <strong>Time</strong> <strong>Series</strong> Prediction.1. IntroductionUncover<strong>in</strong>g cause-effect relationships is central <strong>in</strong> many aspects of everyday life <strong>in</strong> bothhighly <strong>in</strong>dustrialized and develop<strong>in</strong>g countries: what affects our health, the economy,climate changes, world conflicts, and which actions have beneficial effects? Establish<strong>in</strong>gcausality is critical to guid<strong>in</strong>g policy decisions <strong>in</strong> areas <strong>in</strong>clud<strong>in</strong>g medic<strong>in</strong>e andpharmacology, epidemiology, climatology, agriculture, economy, sociology, law enforcement,and manufactur<strong>in</strong>g.One important goal of causal model<strong>in</strong>g is to predict the consequences of given actions,also called <strong>in</strong>terventions, manipulations or experiments. This is fundamentallydifferent from the classical mach<strong>in</strong>e learn<strong>in</strong>g, statistics, or data m<strong>in</strong><strong>in</strong>g sett<strong>in</strong>g, whichfocuses on mak<strong>in</strong>g predictions from observations. Observations imply no manipulationon the system under study whereas actions <strong>in</strong>troduce a disruption <strong>in</strong> the naturalfunction<strong>in</strong>g of the system. In the medical doma<strong>in</strong>, this is the dist<strong>in</strong>ction made between“diagnosis” and “prognosis” (prediction from observations of diseases or disease evolution)and “treatment” (<strong>in</strong>tervention). For <strong>in</strong>stance, smok<strong>in</strong>g and cough<strong>in</strong>g might be bothpredictive of respiratory disease and helpful for diagnosis purposes. However, if smok<strong>in</strong>gis a cause and cough<strong>in</strong>g a consequence, act<strong>in</strong>g on the cause (smok<strong>in</strong>g) can changeyour health status, but not act<strong>in</strong>g on the symptom or consequence (cough<strong>in</strong>g). Thus itc○ 2011 I. Guyon, A. Statnikov & C. Aliferis.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!