FUNNEL: Automatic Mining of Spatially Coevolving Epidemics

download FUNNEL: Automatic Mining of Spatially Coevolving Epidemics

If you can't read please download the document

  • date post

    03-Jan-2016
  • Category

    Documents

  • view

    28
  • download

    2

Embed Size (px)

description

FUNNEL: Automatic Mining of Spatially Coevolving Epidemics. Yasuko Matsubara, Yasushi Sakurai (Kumamoto University) Willem G. van Panhuis (University of Pittsburgh) Christos Faloutsos (CMU). Motivation. Given : Large set of epidemiological data. e.g., Measles cases in the U.S. Linear. - PowerPoint PPT Presentation

Transcript of FUNNEL: Automatic Mining of Spatially Coevolving Epidemics

PowerPoint Presentation

FUNNEL: Automatic Mining of Spatially Coevolving EpidemicsYasuko Matsubara, Yasushi Sakurai (Kumamoto University)Willem G. van Panhuis (University of Pittsburgh)Christos Faloutsos (CMU)SIGKDD 2014Y. Matsubara et al.1

MotivationGiven: Large set of epidemiological data

Y. Matsubara et al.2SIGKDD 2014e.g., Measles cases in the U.S.

Linear(Weekly)http://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

2MotivationGiven: Large set of epidemiological data

Y. Matsubara et al.3SIGKDD 2014e.g., Measles cases in the U.S.

LinearYearly periodicity(Weekly)http://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

3MotivationGiven: Large set of epidemiological data

Y. Matsubara et al.4SIGKDD 2014e.g., Measles cases in the U.S.

LinearYearly periodicityVaccineeffect(Weekly)http://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

4MotivationGiven: Large set of epidemiological data

Y. Matsubara et al.5SIGKDD 2014e.g., Measles cases in the U.S.

LinearYearly periodicityShocks,e.g., 1941Vaccineeffect(Weekly)http://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

5MotivationGiven: Large set of epidemiological data

Y. Matsubara et al.6SIGKDD 2014e.g., Measles cases in the U.S.

LinearGoal: summarize all the epidemic time-series, fully-automatically http://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

6

Data descriptionProject Tycho: infectious diseases in the U.S.Y. Matsubara et al.7SIGKDD 201450 states56 diseasesTime (weekly)

1888(> 125 years)Xhttp://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

7

50 states56 diseasesTime (weekly)

1888(> 125 years)Data descriptionProject Tycho: infectious diseases in the U.S.Y. Matsubara et al.8SIGKDD 2014Element x : # of casese.g., measles, NY, April 1-7, 1931, 4000Xxhttp://www.youtube.com/watch?v=6UV3kRV46Zs&t=49s

8Problem definitionGiven: Tensor X (disease x state x time)

Y. Matsubara et al.9SIGKDD 2014X9Problem definitionGiven: Tensor X (disease x state x time)

Find: Compact description of X, automatically

Y. Matsubara et al.10SIGKDD 2014X

P1P2P3P4P5

MEFUNNELX10Problem definitionGiven: Tensor X (disease x state x time)

Find: Compact description of X, automatically

Y. Matsubara et al.11SIGKDD 2014X

P1P2P3P4P5

MEFUNNELXDiscontinuities

Seasonality

11Problem definitionGiven: Tensor X (disease x state x time)

Find: Compact description of X, automatically

Y. Matsubara et al.12SIGKDD 2014X

P1P2P3P4P5

MEFUNNELXNO magic numbers !

Parameter-free!12RoadmapSIGKDD 201413Y. Matsubara et al.MotivationModeling power of FUNNELOverview main ideasProposed model idea #1 Algorithm idea #2ExperimentsDiscussionConclusionsModeling power of FUNNELY. Matsubara et al.14SIGKDD 2014Questions about epidemicsQ1Q2Q3Q4Q5X14XQuestionsY. Matsubara et al.15SIGKDD 2014Are there any periodicities?If yes, when is the peak season?Q1Q2Q3Q4Q5Q115AnswersY. Matsubara et al.16SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201416Y. Matsubara et al.Seasonality

P1Angle: peak season

Radius: seasonality strength FUNNEL: Polar plot16AnswersY. Matsubara et al.17SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201417Y. Matsubara et al.Seasonality

P1Questions

?17AnswersY. Matsubara et al.18SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201418Y. Matsubara et al.Seasonality

P1Questions

?Q: Does Influenza have seasonality? If yes, when? 18AnswersY. Matsubara et al.19SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201419Y. Matsubara et al.Seasonality

P1Questions

?19AnswersY. Matsubara et al.20SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201420Y. Matsubara et al.

Seasonality

P1Influenza in Feb.Detected by FUNNEL(strong seasonality)Detected!20AnswersY. Matsubara et al.21SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201421Y. Matsubara et al.Seasonality

P1Questions

?21AnswersY. Matsubara et al.22SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201422Y. Matsubara et al.Seasonality

P1Questions

?Q: How about measles ? 22AnswersY. Matsubara et al.23SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201423Y. Matsubara et al.Seasonality

P1Questions

?23AnswersY. Matsubara et al.24SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201424Y. Matsubara et al.

Seasonality

P1Measles(childrens)in springDetected!24AnswersY. Matsubara et al.25SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201425Y. Matsubara et al.Seasonality

P1Questions

?25AnswersY. Matsubara et al.26SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201426Y. Matsubara et al.Seasonality

P1Questions

?Q: Which disease peaks in summer?26AnswersY. Matsubara et al.27SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201427Y. Matsubara et al.Seasonality

P1Questions

?27AnswersY. Matsubara et al.28SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201428Y. Matsubara et al.

Seasonality

P1Detected!Lyme-disease (tick-borne)in summer28AnswersY. Matsubara et al.29SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201429Y. Matsubara et al.Seasonality

P1Questions

?29AnswersY. Matsubara et al.30SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201430Y. Matsubara et al.Seasonality

P1Questions

?Q: Which disease has no periodicity?30AnswersY. Matsubara et al.31SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201431Y. Matsubara et al.Seasonality

P1Questions

?31AnswersY. Matsubara et al.32SIGKDD 2014Q1Q2Q3Q4Q5SIGKDD 201432Y. Matsubara et al.

SeasonalityP1Detected!Gonorrhea(STD)no periodicity32QuestionsY. Matsubara et al.33SIGKDD 2014Can we see any discontinuities?Q1Q2Q3Q4Q5Q2X33AnswersY. Matsubara et al.34SIGKDD 2014Q1Q2Q3Q4Q5Disease reduction effectP2Measles

1965: Detected by FUNNEL1963: Vaccine licensure

Detected!34QuestionsY. Matsubara et al.35SIGKDD 2014Q3Whats the difference between measles in NY and in FL?Q1Q2Q3Q4Q5

35AnswersY. Matsubara et al.36SIGKDD 2014Q1Q2Q3Q4Q5area sensitivity

P3FUNNELs guess of susceptibles (measles)

CATXNY, PA(more children)FL (fewer children)Detected!36XQuestionsY. Matsubara et al.37SIGKDD 2014Q4Are there any external shock events, like wars?Q1Q2Q3Q4Q537

AnswersY. Matsubara et al.38SIGKDD 2014Q1Q2Q3Q4Q5external shock events

P4Funnel can detect external shocksfully-automatically ! World war IIScarlet fever

Detected by FUNNELDetected!38XQuestionsY. Matsubara et al.39SIGKDD 2014Q5How can we remove mistakes and incorrect values?Q1Q2Q3Q4Q5TYPO!39AnswersY. Matsubara et al.40SIGKDD 2014Q1Q2Q3Q4Q5mistakesP5It can also detect typos, automatically !!

Missing valuesTyphoid fever cases

Detected!Mistake40Modeling power of FUNNELOur model can capture 5 properties

Y. Matsubara et al.41SIGKDD 2014P1P2P3P4P5SeasonalityDisease reductionsArea sensitivityExternal eventsMistakes

41RoadmapSIGKDD 201442Y. Matsubara et al.MotivationModeling power of FUNNELOverview main ideasProposed model idea #1 Algorithm idea #2ExperimentsDiscussionConclusionsProblem definitionGiven: Tensor X (disease x state x time)

Find: Compact description of X, automatically

Y. Matsubara et al.43SIGKDD 2014X

P1P2P3P4P5

MEFUNNELX43Two main ideasIdea #1: Grey-box model

Idea #2: MDL for fittingSIGKDD 2014Y. Matsubara et al.44

P1P2P3P4P5

MEFUNNELXNO magic numbers !(parameter-free)Two main ideasIdea #1: Grey-box model - domain knowledge

SIGKDD 2014Y. Matsubara et al.45

P1P2P3P4P5

MEFUNNELX(SIRS+) : 6 parameters

ShocksVaccineTwo main ideasIdea #2: Fitting with MDL -> parameter free!SIGKDD 2014Y. Matsubara et al.46

P1P2P3P4P5

MEFUNNELXNO magic numbers

Parameter-free!

RoadmapSIGKDD 201447Y. Matsubara et al.MotivationModeling power of FUNNELOverview main ideasProposed model idea #1 Algorithm idea #2ExperimentsDiscussionConclusionsProposed model: FUNNELY. Matsubara et al.48SIGKDD 2014statesdiseasesTime

single epidemicMulti-evolving epidemics(a) FUNNEL-single (b) FUNNEL-full

X48Proposed model: FUNNELY. Matsubara et al.49SIGKDD 2014statesdiseasesTime

single epidemicMulti-evolving epidemics(a) FUNNEL-single (b) FUNNEL-full

X49FUNNEL with a single epidemicGiven: single epidemic sequence

Find: nonlinear equation,model parametersY. Matsubara et al.50SIGKDD 2014

e.g., measles in NYFUNNEL50FUNNEL with a single epidemicWith a single epidemic: Funnel-REY. Matsubara et al.51SIGKDD 2014

S(t)I(t)V(t)LinearLogI(t) People of 3 classesS: SusceptibleI : Infected V : Vigilant/ vaccinatedDetails51FUNNEL with a single epidemicWith a single epidemic: Funnel-REY. Matsubara et al.52SIGKDD 2014

S(t) : susceptibleI (t) : Infected V(t) : Vigilant /VaccinatedSVI

Details52FUNNEL with a single epidemicWith a single epidemic: Funnel-REY. Matsubara et al.53SIGKDD 2014

SVI

Details: strength of infection(yearly periodic func)

53FUNNEL with a single epidemicWith a single epidemic: Funnel-REY. Matsubara et al.54SIGKDD 2014

SVI

Details : healing rate : disease reduction effect

54FUNNEL with a single epidemicWith a single epidemic: Funnel-REY. Matsubara et al.55SIGKDD 2014

SVI

Details : temporal susceptible rate

55Proposed model: FUNNELY. Matsubara et al.56SIGKDD 2014statesdiseasesTime

single epidemicMulti-evolving epidemics(a) FUNNEL-single (b) FUNNEL-full

X56

statesdiseases

time

X

(P1, P2): global/country(P3): local/stateM

(P4, P5): extra - E: shocks & M: mistakesE

Proposed model: FUNNEL-fullY. Matsubara et al.57P1P2P3P4P5SIGKDD 2014

statesdiseases

time

X(P1, P2): global/countryProposed model: FUNNEL-fullY. Matsubara et al.58P1P2DetailsGlobalP1P2Base matrix (d x 6)Disease reduction matrix (d x 2)

SIGKDD 2014statesd