Original Investigation
DOI:
https://doi.org/10.30827/Digibug.80900
Performance analysis in tennis since 2000: A systematic review focused on the methods of data collection
Análisis del rendimiento en tenis desde el año 2000: una revisión sistemática enfocada en los métodos de recolección de datos
International Journal of Racket Sports Science, vol. 4(2) (July-Dec, 2022), Pag. 40-55 . eISSN: 2695-4508
Received: 01-04-22
Acepted: 03-03-23
AUTHORS
Hiroo Takahashi
Professor, National Institute of Fitness and Sports in Kanoya, Kagoshima, Japan.
Shuhei Okamura
Assistant Professor, Osaka University of Health and Sports Sciences, Osaka, Japan.
Shunsuke Murakami
Corresponding Author: Hiroo Takahashi,mailto:hiroo@nifs-k.ac.jp
Cite this article as: Jindo, T., Mitsuhashi, D., & Kubota, T. (2022). Accuracy of subjective stats of key performance indicators in tennis. International Journal of Racket Sports Science, 4(2), 40-55.
ABSTRACT
Abstract
In tennis, performance analysis has advanced primarily as notational analysis. And analytical techniques markedly advanced, particularly in the fields of notational analysis and match analysis. In tennis, the Hawk-Eye system was introduced to tour tournaments in 2002. It has recently become used for player tracking and post-match analysis, there are a number of papers using Hawk-Eye data. Along with the development such measuring devices, technologies for analysis of a vast amount of data collected with these devices (big data) have also been developed. In particular, analysis by machine learning using AI was developed in the field of engineering, and it is also increasingly adopted in the field of sports. In the present review, we aimed to clarify the direction of research on performance analysis of tennis by organizing the trend of studies of performance analysis after 2000 with particular attention to the methods of data collection in the hope of furthering the development of this field. As a result of search of reports concerning performance analysis of tennis published after 2000 with particular interest in data collection methods, 90 papers were retrieved. The data collection methods were classified into active and passive methods, and subclassified into categories, i.e., tracking, video recording, data mining, observations of coaches, websites, and broadcasting. This review of the papers in different categories may aid in developing future directions of research in the field of performance analysis in tennis.
Keywords: tracking, coding system, video recording, public data, simulation.
Resumen
En tenis, el análisis del desempeño ha evolucionado principalmente como análisis notacional. Y las técnicas analíticas han avanzado de manera notable, especialmente en los campos del análisis notacional y de partidos. En tenis, el sistema Hawk-Eye fue incorporado a los torneos de circuito en 2002. Recientemente se ha usado para el seguimiento de jugadores y el análisis posterior al partido, y existen diversos artículos que usan datos del Hawk-Eye. Junto con el desarrollo de dichos dispositivos de medición, también se ha desarrollado tecnología para el análisis de grandes cantidades de datos recolectados con estos dispositivos (macrodatos). En particular, se desarrolló en el campo de la ingeniería el análisis con aprendizaje automático e IA, y cada vez es más usado en el ámbito deportivo. En esta revisión, el objetivo fue clarificar la dirección de la investigación sobre el análisis del rendimiento en tenis al organizar la tendencia de los estudios de análisis del rendimiento después del año 2000 con particular atención a los métodos de recolección de datos con el fin de continuar con el desarrollo de este campo. Como resultado de la búsqueda de artículos relacionados con el análisis del rendimiento en tenis publicados después del año 2000 enfocada en métodos de recolección de datos, se encontraron 90 artículos. Los métodos de recolección de datos fueron clasificados en activos y pasivos, y subclasificados en categorías, por ejemplo, seguimiento, grabación de video, minado de datos, observaciones de entrenadores, sitios web y transmisiones. Esta revisión de artículos en diferentes categorías puede ayudar en el desarrollo de otras líneas de investigación futuras en el campo del análisis del rendimiento en tenis.
Palabras clave: seguimiento, sistema de codificación, grabación de video, datos públicos, simulación.
Introduction
INTRODUCTION
Performance analysis is a new concept. Lees (2003) reviewed studies on racket sports by field and selected notational analysis as a category, but did not mention performance analysis.
The paper by Bartlett (2001) is considered to be the first on performance analysis. On the basis of differences between earlier biomechanical studies and studies of notational analysis, Bartlett (2001) defined the value of performance analysis as analysis of good and bad performances of the team and players according to performance indicators used in each genre of studies.
Thereafter, O’Donoghue (2010) defined performance analysis as investigation of sports performance using analytical methods that not only include biomechanics and notational analysis as reported by Bartlett, but also target data collected by physiological and psychological techniques.
In tennis, performance analysis has advanced primarily as notational analysis. As objectives of notational analysis, Hughes (1998) mentioned 1) tactical evaluation; 2) technical evaluation; 3) analysis of movement; 4) development of a database and modelling; and 5) educational use for both coaches and players, and reviews have since been reported according to these 5 goals. O’Donoghue (2004) also wrote reviews using the 5 goals proposed by Hughes (1998), but suggested, as prospects for the future, transformation of match analysis itself with the development of its techniques in addition to the necessity of conducting practical match analysis in the context of coaching.
The development of analytical techniques was previously brought up by Liebermann et al. (2002) . Describing analytical methods for sports performance using the latest IT technology at the time, Liebermann et al. (2002) proposed that these technologies should be utilized in everyday coaching.
These reviews generally targeted papers published before 2000. Thereafter, analytical techniques markedly advanced, particularly in the fields of notational analysis and match analysis. In tennis, the Hawk-Eye system was introduced to tour tournaments in 2002 (hawkeyeinnovations.com, online). The initial objective of this system was to assist line judges, but as it has recently become used for player tracking and post-match analysis, there are a number of papers using Hawk-Eye data . In addition, instruments for tracking of the ball and players, such as Trackman (Trackman Inc.) and PlaySight (PlaySight Interactive ltd.), have been developed, and studies using such instruments are being conducted (Edelmann-Nusser et al., 2019; Murata and Takahashi, 2020: Kashiwagi et al., 2021).
Along with the development such measuring devices, technologies for analysis of a vast amount of data collected with these devices (big data) have also been developed. In particular, analysis by machine learning using AI was developed in the field of engineering, and it is also increasingly adopted in the field of sports. In studies concerning tennis, machine learning has been used by researchers including Whiteside and Reid (2017) , Ganser et al. (2021) , and Fernandes (2017) .
In view of the changes in the methods for collection and analysis of data related to performance analysis of tennis, we considered it necessary to evaluate research themes of future performance analysis of tennis based on a review of papers published after 2000, when such changes became apparent.
In the present review, we aimed to clarify the direction of research on performance analysis of tennis by organizing the trend of studies of performance analysis after 2000 with particular attention to the methods of data collection in the hope of furthering the development of this field.
Methods
METHODS
This review was conducted according to the procedure of systematic review (Pickering and Byrne, 2014). To retrieve the relevant literature, searches were performed with “tennis”, “performance”, “analysis”, “notation”, and “match” as search words with the ‘AND’ condition, excluding “table” and “paddle” to restrict the search to reports concerning tennis. The databases searched were PubMed, Web of Science, and SPORT DISCUS, which encompass the literature concerning sports science. Searches were performed using the above search words in the default mode of each database, by which papers were retrieved if the search words were included in the title, abstract, or keywords. Two additional conditions, i.e., in English and published after 2000, were used for the search. The last date that we searched was April 23rd, 2021.
By the above method, 1,068 papers were retrieved. Of the retrieved papers, those that fulfilled the following conditions were included in the present review: 1) studies of performance in matches, 2) studies aiming to develop analytical methods, 3) studies analyzing quantitative data, and 4) studies published in the category of “research paper” in each journal. Papers that corresponded to the following were excluded as irrelevant to the objective of the present review: 1) studies focusing on physiological, psychological, and/or biomechanical indices alone as analytical targets, and 2) studies focusing on techniques of tennis and/or their development alone. In the first step, all the papers were screened by title, and all the authors agreed on the 130 papers that were retrieved. These papers were screened by Abstract to identify those that fulfilled the inclusion criteria, and all the authors agreed on the 90 papers that were retrieved.
While classifying the retrieved literature, attention was paid to the methods of data collection employed in each study. After overviewing the 90 retrieved papers, they were classified according to the data collection method from the following viewpoints: 1) primary data collection: match data collected using videos and tracking systems at the sites of actual matches or data collected by the researchers themselves using audio-visual media, and analytical data prepared by the researchers themselves by conducting simulations using data from such sources, and 2) secondary data collection: data collected from broad sources, such as those made public online, or that were broadcast on the television. In addition, reports classified into 1) and 2) were subclassified according to the data collection method, and the characteristics of the subclasses were evaluated. The procedures of present review were showed on Figure 1.
Results
RESULTS
According to the data collection methods, 42 and 48 of the 90 papers were classified as using primary and secondary data collection methods, respectively.
1. Studies using primary data collection
Primary data collection methods were subclassified into automatic data collection using tracking technologies, collecting data from video images, handling data from data mining, and collecting data from the observations of coaches. Tracking technologies were classified into vision-based technologies and inertial measurement unit (IMU)-based technologies. Data from video images were classified into automatic and manual methods.
1.1. Studies using tracking technologies
There were 16 studies using vision-based tracking technologies (Table 1-1 and 1-2). In this category, the study using Hawk-Eye data by Loffing et al. (2010) was published earliest. Of the reports evaluated in this review, all those using tracking technologies were published after 2010. Publication of studies using Hawk-Eye data increased, particularly after about 2016 (Kolbinger and Lames, 2013; Mecheri et al., 2016; Reid et al., 2016; Wei et al., 2016; Kovalchik and Albert, 2017; Kovalchik and Reid, 2018; Cui et al., 2019; Meurs et al., 2021). A characteristic of these studies was a large data size. Among the studies using Hawk-Eye data, the study by Mecheri et al. (2016) collected data from more than 100,000 points, and the study by Kovalchik and Albert (2017) targeted more than 30,000 services. They can be regarded as big data analyses.
There were two studies using IMU-based technologies (Tables 1,2). A study by Myers et al. (2019) adopted Sony Smart Sensor, and a study by Edelmann-Nusser et al. (2019) adopted BABOLAT and HEAD sensors. However, as these studies reported that the measurement accuracy of the sensors were inacceptable, the studies that used these sensors were not published.
Table 1-1. Data collection methods of vision-based tracking studies
| Authors | Year | Subject | Methods | Output data |
|---|---|---|---|---|
| Loffing et al. | 2010 | 8098 rallies from 37 men's and 17 women's matches played at ATP, WTA, and Grand Slam | Hawk-Eye | % of the ball placements on opponent's backhand side |
| Kolbinger and Lames | 2013 | 10418 serves of 53 right-handed male players from 56 men’s singles Grand Slam matches of 2010 and 2011on hard court | Hawk-Eye | the placement of the ball of right-handed men's serves |
| Martínez-Gallego et al. Martínez-Gallego et al. Martinez-Gallego et al. | 2013a 2013b 2019 | 188 games in 8 matches recorded at the ATP tournament 500 Valencia in 2011 11 professional players (age 24.8 ± 2.9) ranked between 5 and 113 on the ATP ranking | the SAGIT tracking system | (2013a) distance covered, average speed, time spent in the areas (2013b) % of unforced errors, % of winners and forced errors (2019) time, distance coveres, speed, winners, errors |
| Stare et al. | 2015 | boys U14 (n=11) and girls U14 (n=10) in the national championships in Slovenia ATP tournaments (n=7) | the SAGIT tracking system | the efficiency of the first and second serves the efficiency of the forehand and backhand the efficiency of the forehand and backhand in the return of serve the efficiency of topspin forehand or backhand, the slice of forehand or backhand |
| Mecheri et al. | 2016 | professional tennis tournaments (ATP and WTA) including Grand Slam between 2003 and 2008 75587 points for the women 187009 points for the men | Hawk-Eye | the relationships between the various characteristics of the serve (speed, location, spin, etc) and winning-point probabilities |
| Reid et al. | 2016 | 102 male and 95 female players during the 2012-2014 Australian Open | Hawk-Eye | Serve performance Return of serve performance Groundstroke performance Movement characteristics |
| Wei et al. | 2016 | 8780 shots of the top 3 players (Djokovic, Nadal, Federer) in the 2012 men's Australian Open | Hawk-Eye | Ground stroke speed ratio Ground stroke depth ratio Ground stroke angle ratio Lateral player movement ratio |
| Kovalchik and Albert | 2017 | 175 matches from 2016 Australian open 87 matches of men and 88 matches of women | Hawk-Eye | time-to-serve rally length shot importance |
| Pereira et al. | 2017 | 8 professional players during 4 matches of an international tournament (Futures level) on outdoor clay court in Brazil | Automatic tracking system by Figueroa et al. (2006) | Physical performance Technical performance |
| Kovalchik and Reid | 2018 | 246 matches and 270,023 shots from men and 257 matches and 178,136 shots from women in 2015-2017 Australian open | Hawk-Eye | shot types (clustered by location, shape and speed) % of point won |
| Pereira et al. | 2018 | 10 of U18 players from ITF tournament 8 professional players from Futures 10 professional players from ATP250 | Automatic tracking system by Figueroa et al. (2006) | Time spent of interpersonal coordination patterns during lateral displacements: Anti-phase, In-phase, Serving player phase and Returning player phase |
| Cui et al. | 2019 | 1188 of men, 189 individual players, from four Grand Slam's 2015-2017 | Hawk-Eye | technical-tactical and physical performance |
| Floyd et al. | 2020 | 5 matches from 2015 US Open | no show (only showed as 'tennis player-tracking data') | ESV (Expected Shot Value) |
| Meurs et al. | 2021 | 64 men's matches from 2017 Australian open | Hawk-Eye | PA (Positional Advantage) index by Carvalho et al. (2013) |
The reports in the table are arranged in: 1) chronological order, and 2) alphabetical order using the name of the authors. Studies by the same authors and those using the same methods are unified in the same row.
Table 1-2. Data collection methods of IMU-based tracking studies
1.2. Studies using video images
There were three studies that collected video images by the automatic method (Table 2-1). These studies used an independently developed system that processed video images automatically.
There were 18 studies that collected video images by the manual method (Table 2-2). These studies had been published since 2000. The methods of data collection in this category consisted of two types: observation of video images (Johnson and McHugh, 2006; Jans, 2007; Mergheş et al., 2014; Schmidhofer et al., 2014; Martin-Lorente et al., 2017), and developing independent systems (Klaassen and Magnus, 2003; Hizan et al., 2010, 2011, 2014, 2015; Klaus et al., 2017; Prieto-Lage et al., 2018). Many of studies targeted singles matches from Grand Slam tournaments.
Table 2-1. Data collection methods of video images by automatic system studies
| Authors | Year | Subject | Methods | Output data |
|---|---|---|---|---|
| Connaghan et al. | 2013 | twelve complete matches with players of various skill levels, 825 min in total same as above | Automated tennis event indexing system Match Point: visual coding system | accuracy of event detection user's evaluation |
| Polk et al. | 2014 | two-set match of the best singles players on the coaches' team | TennisVis | the scoreline by Pie Meter View point outcome by Fish Grid View match summary by Filters and Bar Charts |
| Lara et al. | 2018 | a simulated match by two players | comparison of the manual versus automatic tracking | player's positioning |
Table 2-2. Data collection methods of video images by manual system studies
| Authors | Year | Subject | Methods | Output data |
|---|---|---|---|---|
| Klaassen and Magnus Klaassen and Magnus Klaassen and Magnus | 2001 2003 2009 | (2001) 481 matches (male: 258, female: 223) at Wimbledon during 1992-1995 57,319 points in male, 28,979 points in female (2003,2009) all singles matches at Wimbledon 1992-1995 | (2001,2009) no information for data collection (In each match we know the two players and the complete sequence of points.) (2003) TENNISPROB | (2001) dynamic binary panel data with random effects tests whether points in professional tennis are iid (independent and identically distributed) (2003) forecasting the probability of winning a match (2009) the efficiency of winning a point on serve |
| Johnson and McHugh | 2006 | 22 players on 3 Grand Slams (8 in RG, 11 in Wimbledon, 9 in US) in 2003 | observation from video recording | number of strokes stroke distribution |
| Jans | 2007 | 3 final matches from 3 Grand Slams (RG, Wimbledon, US) in 2005 | observation from video recording | time duration of point time interval of point total time of match time of play |
| Hizan et al. | 2010 | tennis coding system | coded the same match on two occasions separated by a 4-week period 5 raters coded 674 shots | intra-rater reliability inter-rater reliability comparison with Hawk-Eye data |
| Hizan et al. Hizan et al. Hizan et al. | 2011 2014 2015 | (2011) 28 matches (male:14, female: 14) from 2008 Australian Open, 2666 points (male: 1651, female: 1015) 28 U-16 (male: 14, female: 14) matches and 28 U-12 (male: 14, female: 4) matches from 2008 Australian Boys and Girls championships, 2359 points on U-16 (male: 1239, female: 1120) and 2267 points on U-12 (male: 1175, female: 1092) (2014) 23 matches (male:11, female: 12) from 2008 Australian Open, 1968 successful serves (male:1172, female: 796) 27 U-16 (male: 14, female: 13) matches and 21 U-12 (male: 12, female: 9) matches from 2008 Australian Boys and Girls championships, 2836 succesful serves on U-16 (male: 1439, female: 1397) and 1647 succesful serves on U-12 (male: 916, female: 731) (2015) 23 matches (male:11, female: 12) from 2008 Australian Open, 5221 serves (male: 3272, female: 1949) 27 U-16 (male: 14, female: 13) matches and 21 U-12 (male: 12, female: 9) matches from 2008 Australian Boys and Girls championships, 3391 serves on U-16 (male: 1740, female: 1651) and 1922 serves on U-12 (male: 1050, female: 872) | tennis coding system (by Hizan et al., 2010) | (2011) % 1st in, aces, DF, % 1st won, % 2nd won, % 1st return won, % 2nd return won (2014) serve-return location point winning (2015) serve location point winning |
| Carvalho et al. Carvalho et al. | 2013 2014 | (2013) 27 rallies in 3 matches from 2008 Estoril Open (ATP 250) (2014) 28 rallies in 3 matches from 2008 Estoril Open (ATP 250) | recording by DV camera and 2D-DLT | (2013) PA (Positional Advantage) index (2014) GDD (Goal-Directed Displacement) index |
| Mergheş et al. | 2014 | 9 matches by 3 players (Federer, Nadal, Agassi) in 2 years | observation from video recording | % won on 1st serve % won on 2nd serve % won on return |
| Schmidhofer et al. | 2014 | 12 matches each for 3 groups (U9, U10, U12) from Develop Tournaments for Australian Tennis Association 12 matches of ATP tournaments | observation from video recording | service parameters return parameters ICT (Inter Contact Times) parameters miscellaneous parameters |
| Fitzpatrick et al. | 2017 | 48 participants MTR: n=18, Age 7.4 ± 0.6, 230 points MTO: n=16, Age 8.5 ± 0.6, 253 points MTG: n=8, Age 9.9 ± 0.4, 280 points FB: n=6, Age 13.7 ± 0.5 247 points | a custom-notational analysis system | rally length shot type shot variety winners and errors serves |
| Klaus et al. | 2017 | 8 U-14 national level male players in Australia QF and SF of the Victorian Junior Hardcourt Championships | A developed computerized system Kinovea (version 0.8.15) | type of stroke type of outcome court position |
| Martin-Lorente et al. | 2017 | 18 matches of Grand Slam and ATP finals between 2011 and 2014 11 men players | observation from video recording | results of inside out and inside in forehand |
| Prieto-Lage et al. | 2018 | 82 break point events between Nadal and Djokovic on final clay court during 2011 and 2012 | observation from video recording with OBSTENNIS | the break points T-Pattern |
| Martínez-Gallego et al. | 2021 | 2339 points from 19 complete doubles matches of the 2018 ATP World Tour Masters 1000 tournament played in Canada | a data collection system was designed using Microsoft Excel | time characteristics of doubles tennis time characteristics of the points by winning and losing team time characteristics of the points by the type of match |
1.3. Studies using data mining
There were two studies using data mining theory (Table 3). These studies aimed to predict the results of matches or simulate the progression of matches.
Table 3. Data collection methods of data mining studies
| Authors | Year | Subject | Methods | Output data |
|---|---|---|---|---|
| O'Donoghue and Simmonds | 2019 | Traditional tennis games Traditional tiebreaks Fast4 tennis games Tiebreaks in Fast4 tennis Tiebreak Ten | Simulation in various winning point probabilities | The probability of the player who serve first |
| Li et al. | 2021 | no information | data mining technology | serve points won and lost |
1.4. Studies using data from the observations of coaches
A study that aimed to clarify performance analysis in tennis using data of the observations of coaches (Torres-Luque et al., 2018) was classified into this category (Table 4).
Table 4. Data collection methods of studies using the observations of coaches
Studies of secondary data collection
Secondary data collection methods were subclassified into data collection from official websites and data collection from video images published by television broadcasting and websites.
2.1. Studies using data collected from websites
There were 38 studies using information released on websites (Table 5). Such studies were more common after 2010. Most of the studies targeted men’s singles matches and collected data from the official ATP website and official Grand Slam website. Some studies targeted women’s, doubles, and junior matches (Brenzik, 2013; Kovalchik et al., 2017; Cui et al., 2018; Sogut, 2018; Fernandez-Garcia et al., 2020; Li et al., 2020; Grambow et al., 2021). Other studies collected data from websites that gathered match data independently (Kovalchik and Reid, 2017; Kovalchik and Ingram, 2018; Fagan et al., 2019; Ingram, 2019; Makino et al., 2020). In addition, there were some studies that had no information about the data source (Pollard et al., 2006; Newton and Aslam, 2009; Tudor et al., 2014; Gu and Saaty, 2019; Stefani, 2020). A characteristic of these studies was their large data size.
Table 5. Data collection methods of public data on websites
| Authors | Year | Subject | Source | Output data |
|---|---|---|---|---|
| Pollard et al. | 2006 | 4883 matches data from 1995-2004 All Grand Slam tournaments of men's singles | no show | the probability of winning a set in a match iid (independent and identically distributed) in a set |
| Djurovic et al. | 2009 | 128 matches data from 2007 and 2008 Grand Slam hard court tournaments from IBM | IBM DB2 applicaction | the latent (factor) area of a tennis match |
| Newton and Aslam | 2009 | 330 players over 59 ATP tournaments in 2007 four Grand Slams four top players (Federer, Nadal, Roddick, Blake) | no show | percentage of points won on serve percentage of points won on receiving serve Monte Carlo simulations |
| Reid et al. | 2010 | 2007 Matchfact information of the top 100 male professional players | ATP website | correlation coefficients between the different performance variables |
| O'Donoghue | 2012 | Study 1: 92 men's singles matches in the 2011 US Open from the official website Study 2: world top four players in Grand Slam tournament between 2008 and 2011 | Study 1: 2011 US Open website Study 2: official Grand Slam website | expected and observed break points per receiving game probability of winning points during receiving game |
| Breznik | 2013 | male (N=16,732) and female (N=16,432) players between 1968 to 2011 obtained from ATP and WTA website | ATP website WTA website | number of matches won by handedness results of PageRank algorithm |
| Ma et al. | 2013 | 18,288 performances between 1991 and 2008 from the website of the ATP | ATP website | predicting winner or loser by logistic regression model with three variables (match characteristics, personal characteristics, skills and performance) |
| O'Donoghue | 2013 | men's and women's matches from 2012 Grand Slam tournaments | official Grand Slam website | propotion of points won on serving probability of rare events occurred |
| Vaverka and Cernosek | 2013 | players participated in all four Grand Slam in 2008 | official yearbooks and website of the ITF official Grand Slam website | correlation coefficients between body height and serve speed |
| Bane et al. | 2014 | rankings data and date of birth information from 1985 to 2010 | ATP website | the age of first ATP ranking the time to reach Top 100 from first ranked the time between first entry and exit from Top 100 the time between first ATP ranked and exit from Top 100 |
| Kovalchik | 2014 | 498 competitors in end-of-year ATP rankings of 104 or higher between 1991 and 2012 | ATP website | trends in player characteristics (30 and over, teenagers, the age of peak performance, etc) with local polynomial regression curves |
| Tudor et al. | 2014 | all the matches from main draws of Roland Garross, Wimbledon and US Open in 2010 and 2011 | no show | match statistics |
| Filipcic et al. | 2015 | male players ranked in top 300 on ATP ranking in 1991, 2000 and 2010 match statistics of 1961 matches from 1991, 2363 matches from 2000, 2660 matches from 2010 | ATP website | match statistics |
| Kim et al. | 2015 | 2012 Australian Open SF video on the web a men's match and a women's match | video on a website (site information is no show) A coordinate system by Matlab | location of the ball bounce time series of ball anble differences |
| Kovalchik | 2016 | 53,442 matches played by ATP top 100 players in 2004-2014 and 1,377 matches from 2015 | ATP website | fitted model of the Pythagorean theorem |
| Prieto-Bermejo et al. | 2016 | ATP top 10 players on four Grand Slams between 1990 and 2012 | ATP website | relationships between ranking position and the results on tournaments |
| Kovalchik and Reid | 2017 | match activity from 2000-2015 of junior players from ITF website professional men's and women's players from Tennis Abstract website point-by-point data for Grand Slam matches from FlashScore website more detailed point-by-point data from Hawk-Eye | ITF website Tennis Abstract website FlashScore website Hawk-Eye | relative importance of match statistics for winning |
| Kovalchik et al. | 2017 | 877 player trajectories entered WTA rankings between 1989 and 2016 | WTA website | the mean peak ranking in the first ranking year, the number of years during which the majority of progression occurred (the progression stage), and the rate of rankings gained during the progression stage |
| Cui et al. Cui et al. | 2017 2020b | 1188 players in 594 matches collected from four 2015-2017 Grand Slams men's singles | official Grand Slam website | relationships between match statistics and relative quality (RQ) difference of performance indicators between seeded players and non-seeded players |
| Cui et al. | 2018 | 1369 matches in four Grand Slams women's singles | official Grand Slam website | relationships between match variables and the relative quality (RQ) performance profiles |
| Kovalchik and Ingram | 2018 | 1582 men's matches and 966 women's matches from 2010 to the present 33,788 points across 161 men’s matches and 21,450 points across 170 women’s matches at the 2015 and 2016 Australian Opens by Hawk-Eye | the Match Charting Project (www.tennisabstract.com) Hawk-Eye | point distribution by match format time distribution by match format impact of match format on match durations and upsets |
| Sogut | 2018 | male (n=60) and female (n=59) players in 2017 Wimbledon | ATP website WTA website | correlation between body height and mtatch outcomes |
| Vaverka et al. | 2018 | men (n=72-92) and women (n=70-98) at four Grand Slams in 2008, 2012 and 2016 | official Grand Slam website | differences in the serve speed of Grand Slams |
| Fagan et al. | 2019 | handedness data as well as match- play results from ATP Tennis in 2014 | ATP Tennis Navigator (http://www.tennisnavigator.com/) | the advantage of left-handedness probability of match-play results |
| Fitzpatrick et al. | 2019a | 244 men's matches and 250 women's matches from 2016 and 2017 French Open | 2016 and 2017 Roland Garros website | relationships between performance characteristics and PWOL (Percentage of matches in which the Winner Outscored the Loser) |
| Fitzpatrick et al. | 2019b | 244 men's matches and 250 women's matches from 2016 and 2017 French Open 241 men's matches and 249 women's matches from 2016 and 2017and Wimbledon | Roland Garros website and the Wimbledon information System by IBM | relationships between performance characteristics and PWOL (Percentage of matches in which the Winner Outscored the Loser) |
| Gu and Saaty | 2019 | 82987 matches from 1990 for ATP and 35886 matches from 2003 for WTA | online sites | predicted the outcome of 2015 US OPEN |
| Ingram | 2019 | 2208 matches from ATP 2014 season | MatchStat.com (scraping) | a point-based Bayesian hierarchical model for predicting the outcome of tennis matches (the probability of winning a point on serve given surface, tournament and match date) |
| Martin et al. | 2019 | 50 five-set matches from 2014 Grand Slams | official Grand Slam website | effect of pacing strategies on match outcome effect of players’ ATP ranking on pacing strategies effect of Grand Slam tournament on pacing strategies |
| Cui et al. | 2020a | 146 men's matches from 2016-2017 US Open and Australian Open | official website of each tournament | set-to-set differences of match performance |
| Damani et al. | 2020 | 127 men's matches from 2020 Australian Open | 2020 Australian Open website | differences of match statistics among entire tournament, initial rounds (1R-4R) and intense rounds (QF, SF and F) |
| Fernandez-Garcia et al. | 2020 | 546 matches by professionals and U-18 in three Grand Slams | official Grand Slam (Australian Open, Roland Garros and Wimbledon) website | differences of match statistics between professionals and U-18 players |
| Grambow et al. | 2020 | 1772 men's matches from 2002-2015 Wimbledon | Wimbledon information System by IBM | serve performance comparisons by tournaments year and tournament week |
| Li et al. | 2020 | professional players of mens (n=180) and womens (n=193) within top 300 ranking between 2010 and 2018 | ATP website WTA website | relationships between the age and their ranking milestones |
| Makino et al. | 2020 | 4230 points on three surfaces (Hard, Clay, Grass) of four players (Federer, Nadal, Murray, Djokovic) | Match Charting Project (https://github.com/JeffSackmann/tennis_MatchChartingProject) | match winner predictions using machine learning |
| Stefani et al. | 2020 | almost 5000 men's and 5000 women's matches of four Grand Slams from 2006-2019 | no show | percent of matches by the higher-seeded players |
| Grambow et al. | 2021 | 1771 ladies' matches from 2002-2015 Wimbledon | Wimbledon information System by IBM | serve performance comparisons by tournaments year and tournament week |
2.2. Studies using data collected from broadcasting
There were 10 studies using data collected from broadcasting (Table 6). The studies collecting data from terrestrial and satellite broadcasting were published between 2000 and 2012 (O’Donoghue, 2001; O’Donoghue and Ingram, 2001; Gillet et al., 2009; Yu et al., 2009; Nowak and Panfil, 2012), whereas recent studies collected video images from websites (Carboch et al., 2018a, 2018b, 2019, 2020; Martinez-Gallego et al., 2020). Most of the studies targeted singles matches of Grand Slam tournaments, and one study targeted doubles matches (Martinez-Gallego et al., 2020).
Table 6 Data collection methods of broadcasting studies
| Authors | Year | Subject | Methods | Output data |
|---|---|---|---|---|
| O'Donoghue | 2001 | men's and women's 252 matches from Grand Slam tounaments between 1997 and 1999 from terrestrial and satellite television coverage | a computerized data management system | proportion of points won when serving proportion of games won |
| O'Donoghue and Ingram | 2001 | men's and women's 175 matches from Grand Slam tounaments between 1997 and 1999 from terrestrial and satellite television coverage | a specially designed computerized notational analysis system for tennis | differencese of timing factors and strategy data among tournaments and gender |
| Gillet et al. | 2009 | 116 men's matches from French Grand Slam tournament in 2005 and 2006 from terrestrial television coverage | a computerized notational system | serve characteristics and point winning serve-return characteristics and point winning |
| Yu et al. | 2009 | broadcast tennis video | a frame grouping technique | 3D virtual content insertion application ball detection and tracking application |
| Nowak and Panfil | 2012 | the match by Federer and Djokovic of 2007 US Open final and 2008 Australian Open semi-final from broadcasts by Eurosport | data recorded with Microsoft Excel | relationships among type of shot, ball placement on court and fixed or dynamic elements of play |
| Carboch et al. Carboch et al. Carboch et al. | 2018a 2018b 2019 | 23 women's matches from 2017 Australian Open 7 men's and 23 women's matches from 2017 Australian Open 24 men's matches from Austrarian Open, French Open and Wimbledon in 2017 from television or internet broadcast | a spreadsheet for observed variables | comparisons of point duration, number of rally shots, time between the points, rally pace and work to rest ratio |
| Carboch et al. | 2020 | 23 women's matches from 2017 Australian Open and 24 men's matches from Austrarian Open, French Open and Wimbledon in 2017 from television or internet broadcast | a spreadsheet for observed variables | comparisons of match characteristics between new and used balls |
| Martínez-Gallego et al. | 2020 | 34 men's doubles matches from ATP tournaments in 2018 from Tennistv.com | a registration system created with Microsoft Excel | point ending situations |
DISCUSSION
DISCUSSION
1. Studies using primary data collection
Methods using automated vision-based tracking techniques, mainly Hawk-Eye, will continue to be the mainstay of primary data collection. Concerning studies using Hawk-Eye data, a group participated in by Tennis Australia has recently been active in reporting Australian Open matches (Reid et al., 2016; Wei et al., 2016; Kovalchik and Albert, 2017; Kovalchik and Reid, 2018; Meurs et al., 2021). There have also been studies focusing on other tournaments (Loffing et al., 2010; Kolbinger and Lames, 2013; Mecheri et al., 2016; Cui et al., 2019) and expansion of the research field was confirmed. The Hawk-Eye system is routinely employed in major tournaments. Groups conducting these studies reached an agreement with the tournament organizers about the use of the data obtained in the tournaments by the Hawk-Eye system for research. Building such relationships between tournament organizers and Hawk-Eye providers is considered a process indispensable for the development of research in this field.
As mentioned above, there was a limitation to the use of the Hawk-Eye system data; thus, video images obtained by the manual method were used. The manual method of collecting video images was a general methodology. Especially the studies that targeted junior matches, such as those without the Hawk-Eye system, adopted the manual method to collect video images (Schmidhofer et al., 2014; Fitzpatrick et al., 2017; Klaus et al., 2017). Recently, advances in image processing have made it easier to calculate parameters from images obtained with video cameras than before 2000. As the use of video images is a relatively simple method to collect data in environments where it is difficult to employ a high-tech system, such as Hawk-Eye, the use of methods currently employed in other sports events and image processing techniques used in other fields as well as developing original systems for automatic collection of parameters from video images appropriate for the objective of the study using existing techniques as references may be solutions for the establishment of a method for data collection from video images.
We confirmed that two studies used data mining theory (O’Donoghue and Simmonds, 2019; Li et al., 2021). As mentioned below, there were many studies that used published data on the Internet. The field of data mining was prospected to develop a technique for predicting or simulating the results of matches with published data on the Internet.
2. Studies of secondary data collection
Many studies of secondary data collection were carried out by collecting data from websites. On the present website of the ATP Tour (ATP TOUR.com, online), a wide variety of data, including the summary of points scored and the decisive shot at each score called MATCH BEATS, detailed results of rallies called RALLY ANALYSIS, and, on the page called the second-screen, positions where the ball was hit, positions where the ball fell, distance run, and speed of the ball hit, in addition to conventional stats, such as the first-service percentage and first-service scoring rate, are provided. Such detailed data has the same quality as the vision-based tracking data described in this review, and proceeding with exploratory research using such open data may lead to further development of research in the field of performance analysis in tennis. In particular, many studies analyzing such data from a long-time perspective have been conducted, and they are expected to provide findings that will aid in the 4) development of a database and modelling, and 5) educational use for both coaches and players among the 5 viewpoints suggested by Hughes (1998) by making studies from both cross-sectional and longitudinal viewpoints possible.
However, public data from tournaments and matches are limited, and only data of particular tournaments are available. In addition, it was only after 1991 that stats began to be provided and after 2018 that detailed stats began to be released. Therefore, caution is needed in the use of data.
Recently, data collected from broadcasting have become available on websites as streaming services. Data collected by such methods will continue to be used for research.
Most of the studies by secondary data collection targeted men’s singles matches of world top-ranked players. There were few studies of female players, doubles matches, and junior players. It is necessary to perform studies to obtain data about these categories. As mentioned below, studies that targeted junior matches, such as those that were played without the Hawk-Eye system, adopted the manual method to collect video images, especially data from online streaming video for doubles matches.
Concluding
CONCLUSIONS
As a result of search of reports concerning performance analysis of tennis published after 2000 with particular interest in data collection methods, 90 papers were retrieved. The data collection methods were classified into primary and secondary methods, and subclassified into 6 categories, i.e., tracking, video recording, data mining, the observations of coaches, Internet, and broadcasting. This review of the studies in different categories suggests the importance of considering vision-based tracking technologies, the increased use of manual video-recordings, the possibility of data mining, the use of official websites, and performing studies focusing on female players, doubles teams, and junior players.
References