Basketball Data Science - WU PDF Free Download

2y ago

36 Views

1 Downloads

5.86 MB

48 Pages

Report/dmca

Download PDF

Transcription

Basketball data scienceMarica Manisera – Paola ZuccolottoUniversity of Brescia, ItalyVienna, April 13, 2018Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

arica ManiseraPaola Zuccolotto- University of Brescia, Italy

BDSports, a network ofpeople interested inSports Analyticshttp://bodai.unibs.it/bdsports/Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisAgenda: Basketball analytics: state of the art Basketball datasets Case studies:CS1: new positions in basketballCS2: scoring probability when shooting under high-pressure conditionsCS3: performance variability and teamwork assessmentCS4: sensor data analysis Concluding remarksMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisOur analyses oftenintegrate machinelearning tools andexperts’ suggestionsMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Scientific Literature CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisScientific Journals Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketballMarica ManiseraPaola ZuccolottoSpecialIssues- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisDataBig DataSensorDataCS4play-by-playCS2 – CS3Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisCS1: new positions in basketballEJASA Special issueon Statisticsin SportMotivation: The existing positions - often defined a longtime ago - tend to reflect traditional points of view aboutthe game and sometimes they are no longer well-suited tothe new concepts arisen with the evolution of the way ofplaying.Aim: describing new roles of players during the game, by means of the analysisof players' performance statistics with data mining and machine learning tools.Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysis «Key-players» training set 7-dimensional SOM clusterization of the SOMoutput layer into a propernumber of groups bymeans of a fuzzy clusteringalgorithmMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisCS2: scoring probability when shootingunder high-pressure conditionsInternationalJournal ofSports Science& Coachingplay-by-playMotivation: Basketball players have often to facehigh-pressure game conditions. To be aware of theoverall and personal reactions to these situations isof primary importance to coaches.Aim: To develop a model describing the impact of some high-pressure gamesituations on the probability of scoring and to assess players' personal reactions.Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisHigh-Pressure Game Situations: when the shot clock is going to expire (SHOT.CLOCK) when the score difference with respect to the opponentis small (SC.DIFF) when the team, for some reason, has globally performedbad during the match, up to the considered moment(MISS.T) when the player missed the previous shot (MISS.PL) the time to the end of quarter (TIME) type of action (POSS.TYPE, 24’’ or 14’’ extratime)Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball69688 CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysis6470Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisData Mining Tools: univariate non parametric regressions via kernelsmoothing on the dependent variable MADE (assumingvalues 1 and 0 according to whether, respectively, theattempted shot scored a basket or not) 1000 bootstrap samples of size nboot 5000 andnboot 1000 for the dataset A2ITA and RIO16,respectively.few univariate relationships detected - Just SHOT.CLOK and MISS.PLMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisData Mining Tools: CART (Classification And Regression Trees), algorithm able todeal with multivariate complex relationships, also detectinginteractions among predictors we transform numerical into categorical covariates in order toimprove interpretability combination of the results of amachine learning procedure and experts' suggestions pruningMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisfocus on:(1) the last 1-2 seconds of possession, veryclose to the shot clock buzzersounding,(2) games where the score difference islow, for example, between -4 and 4,(3) the last 1-2 minutes of each quarter(especially the final quarter)Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisNew Shooting Performance Measure:Takes into account that shots attempted in differentmoments have different scoring probabilitiesPerformance of Player ifor shot type T(2P, 3P, FT)j-th shot made (1)or missed (0)Marica ManiseraPaola Zuccolottoscoring probabilityof j-th shotaccording to CART- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisFurther Research:according to psychological studies, some athletes view the competitive situationsas challenging, and others perceive the same situations as stressful and anxietyprovoking. For this reason, it may be difficult to statistically detect stressfulsituations from large datasets including several players, as the overall averageperformance may remain unchanged as a response to some players improvingtheir performance and some others getting worse.Analysis of single players’ reactions to stressful gamesituations (propension to shot and variation in scoringprobability)Integration with psychological studiesMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisCS3: performance variability and teamwork(in progress)play-by-playMotivation: Psychological studies have pointed outthat typical performance is but one attribute ofperformance, but other aspects should be taken intoaccount, in particular performance variability.Aim: Assessment of players' shooting performance variability and investigationof its relationships with the team composition.Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisPerformance Variability: Definition of a performance index based on the % ofattempted shots that scored a basket and on theshooting intensityshooting intensityshooting performanceMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisPerformance Variability: Fit Markov Switching models to the shootingperformance index, in order to detect the (significant)presence of periods of good and bad performanceMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisTeamwork Assessment: determine influence of each teammate on the regimeof good and bad performance display the significantrelationships by means ofgraphical network analysis tools predict the best substitutionat a given timeMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisCS4: sensor data analysisMathSport International 2017SIS2017Aim: A first approach to sensor data analysis inbasketball (visualization tools, cluster analysis, futurechallenges)In collaboration with MYagonismMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisCS4: sensor data analysisSIS 2018COMPSTAT 2018Aim: to study structural changes in the surface area associations between regimes and game variables the relation between the regime probabilities andthe scored pointsMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisVisualization ToolsA tool to display data recorded by tracking systemsproducing spatio-temporal traces of player trajectorieswith high definition and frequencyhttps://www.youtube.com/watch?v aejyrDnqYVYMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisVisualization ToolsJames P. CurleyCurley Social Neurobiology Lab website(Psychology Department and Center forIntegrative Animal Behavior, ColumbiaUniversity, New York City)Marica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisCluster Analysis MultiDimensional ScalingMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisFuture Challenges: Integration with play-by-play data Integration with video and match analysis Integration with body metrics (bodyphysiology tracking via “smart clothing”and/or body measurements) Integration with qualitative assessments Network analysis tools Spatio-temporal statistical models Addition of the other team’s data Addition of the ball’s positionMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketballTrue: CS2: scoring probability under high-pressure CS3: performance variability and teamwork CS4: sensor data analysisConcluding If people keepthinking thatStatistics is merelyPPG, AST, REB, If people don’tlearn how Statshave to beinterpreted (“Donot put your faithin what statisticssay until you havecarefullyconsidered whatthey do not say.”W. W. Watt)False: If modernapproaches tobasketballanalytics are used If we are able tointegrate analyticsand technicalexperience If we are able tospread the cultureof StatisticsMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

ReferencesDownload a (regularly updated) list of references nk YouMarica ManiseraPaola Zuccolotto- University of Brescia, Italy

Data Big Data play-by-play CS2 –CS3 Sensor Data CS4 Basketball Analytics: state of the art Basketball datasets CS1: new positions in basketball CS2: scoring probability under high-pressure CS3: performance variability