2024年美赛35篇特等奖O奖论文-C-2401298.pdf

ProblemChosen2024TeamControlNumber

MCM/ICM

C2401298

SummarySheet

MomentumExistsInTennisGameAsResidualEffect-

ADual-TemporalBayesianNetworkModel

Summary

Thisstudyinvestigatestennismatchdynamicsthroughsophisticatedstatisticalmod-

elingandanovelDual-TemporalBayesianNetworkapproach.ByanalyzingWimble-

dondata,weaddressfivekeyquestions:

PerformanceMetrics(Problem1):Weintroduceaserver/returnerreweighting

strategytoaccuratelyevaluatetheperformanceofaplayer.Wefurtherutilizesliding

windowandAreaUnderCurve(AUC)methodstoensurecontinuityandlocality,thus

bettercapturingthefluctuationswithinthegame.

MomentumExistence(Problem2):Rigoroushypothesistesting(Ljung-BoxQTest

andRunsTest)failstorejectnullhypothesisofcompleterandomness.Butinstanta-

neouswinningratescanrevealslightdeviationsfromrandomness,suggestingthe

presenceofsubtlemomentumeffects,albeitnotoverwhelminglyconclusive.

MomentumPrediction(Problem3):Wetakea2-stepapproach.Wefirstseparate

outthemomentumeffectbyconsideringitastheresidualofaNaiveBinomialModel.

Subsequently,wedevelopaDual-TemporalBayesianNetworkModeltopredictthis

residualeffect.Thenetworkincorporatesvariouslatentvariablessuchasphysiology,

perception,andself-efficacy.Anadditionaloutcomeofthismodelisthecalculationof

theimportanceoffactorsonmomentumthroughthereductionofinformationentropy.

PredictiveAnalysis(Problem4):OurmodelistestedontheWimbledon2023com-

petition,whereitsuccessfullypredictsmostmatchfluctuations.Weanalyzeinstances

offailureand