skip to main content
10.1145/3696952.3696982acmotherconferencesArticle/Chapter ViewFull TextPublication PagesiciipConference Proceedingsconference-collections
research-article
Open access

Data-Driven Tennis Strategy Evaluation through Hierarchical Markov Models

Published: 21 November 2024 Publication History

Abstract

This study introduces a novel methodology for evaluating tennis player performance through the application of Hierarchical Markov Models (HMM). By dissecting the game's flow and modeling the transitions between states, our proposed model captures the intricate interplay of short-term and long-term dependencies that characterize match dynamics. Utilizing the HMM framework, we model the scoring sequences and transitions across various levels of the game, ranging from individual points to entire matches. Our analysis of match data unveils a nuanced perspective on player performance, pinpointing critical junctures within matches and offering a visual representation of match progression through directed graphs. The findings demonstrate that HMMs are adept at encapsulating the complexities of tennis matches, thereby providing fresh insights into player strategies and performance metrics.

1 Introduction

Along with the advent of the era of big data and artificial intelligence, data analysis has become increasingly prevalent in sports events [1]. In the realm of football, a vast amount of video data is used for match analysis, aiding coaches and players in formulating tactics and training plans [2]. Through the analysis of match video data, key information such as player movement trajectories, passing routes, and shooting positions can be obtained, providing scientific guidance and recommendations for the team. Golden State Warriors were among the earliest teams to leverage data analysis to enhance their win rate[3]. They analyzed player performance metrics such as scoring, rebounds, assists, and more to assess player contributions and optimize team rosters and tactics. Additionally, data analysis can help teams identify weaknesses and vulnerabilities in their opponents, allowing for the development of more targeted game strategies [4].
In summary, in the era of big data and artificial intelligence, data analysis has become an indispensable component of sports events [5]. It not only helps teams improve their win rates but also offers fans deeper insights into matches and enhances their viewing experience [6]. However, most of the previous experiences that have been executed well involve post-match data analysis [7], and the application of real-time analysis during actual matches is still limited.
This study introduces a HMM to bridge this gap. The HMM framework is capable of modeling the scoring sequences and state transitions at multiple levels of the game, from individual points to entire matches. This approach not only improves the accuracy of performance evaluation but also provides a visual representation of match flow, making it easier to identify key moments and turning points.

2 Related Works

Despite extensive research on momentum in sports, limitations persist, such as a focus on specific scenarios and a lack of real-time predictive capabilities. This study aims to develop a comprehensive momentum assessment model using LSTM and Markov processes for more accurate in-match momentum prediction.
Several researchers have made notable contributions to this field by utilizing Hierarchical Markov Models (HMM) in their studies. Chan, Fearing, Fernandes, and Kovalchik [8] proposed a novel modeling framework based on Markov processes to investigate the impact of execution error on a player's value function and strategy in tennis. Sim and Choi[9] analyzed the winning probability of tennis matches and emphasized the importance of points in determining game outcomes. Gollub[10] explored new methods for forecasting serve performance in professional tennis matches, enhancing the accuracy of match result predictions. Yu, Sun, and Hu[11] introduced a visual analysis of athlete performance based on Markov chains and Exponential Weighted Moving Averages (EWMs), providing insights into performance evaluation in sports. Furthermore, Tea and Swartz[12] conducted a comprehensive analysis of serve decisions in tennis using Bayesian hierarchical models, shedding light on individual player tendencies and match dynamics. Fernandes[13] developed a modeling framework based on Markov processes to investigate the impact of execution error on player performance, revealing insights into optimal shot selection strategies and their effectiveness. These studies collectively contribute to a deeper understanding of tennis match dynamics, player strategies, and performance evaluation. By utilizing HMM and related methodologies, researchers have advanced our knowledge of tennis analytics, enabling more accurate predictions and informed decision-making in the sport [14]. Despite these advancements, there remains a need for models that can comprehensively capture both the short-term and long-term dependencies in match dynamics.
Momentum is defined by Tod [15] as a psychological and physiological state marked by a sequence of sustained successes or failures. Its significance in athletic performance is underscored by the regulatory role of emotions.
The existence of momentum is a contentious issue, with studies such as Morgulev et al. [16] finding no evidence in basketball, while others like Dietl and Nesseler[17] identify a significant impact in tennis.
Psychological momentum significantly influences match results, particularly in scenarios with high emotional investment and frequent critical events, as demonstrated by Iso-Ahola and Dotson [18]. Researchers have employed mathematical models and machine learning to forecast momentum shifts, with Han et al. [19] utilizing Markov chains and random forests in tennis.
Technological advancements have broadened the scope of momentum prediction, with Koski et al. [20] exploring the use of GNN and Zhang et al.[21] proposing an attention-based LSTM model. The integration of machine learning and deep learning in sports performance analysis, as reviewed by Bunker and Susnjak [22] and Alzubaidi et al.[23], indicates the potential for interdisciplinary insights.

3 Experimental Preparation

We use tow open dataset, including Wimbledon featured maches1 and MCM data set2 for the competition:
Browsing through the data ETL, we find that there are 752 NAs, i.e. missing values, for “speed_mph” in the dataset Wimbledon. To ensure the integrity and reliability of the data, we use the nearest neighbor interpolation method to interpolate the missing values. serve_width and serve_depth columns have 54 missing values each, and return_depth column has 1309 missing values. Since they are all typed data, we use the plurality to fill in the missing values. The variables serve_width, serve_depth, and return_depth were then converted to numeric data for use in subsequent models. To reduce the effect of different magnitudes, we normalized p1_distance_run, p2_distance_run, and speed_mph.
In this work, we use the nomenclature in Table 1 in the model construction. Other non-frequent-used symbols will be introduced once they are used.
Table 1:
SymbolsDefinition
SkPlayer k's score in set
gkPlayer k's score in game
pkPlayer k's score in points
FijProbability that player i serves to win
PmProbability of winning a match
Table 1: Notation Descriptions

4 Tracking the scoring process of a game

The hierarchical Markov model is employed to track the scoring process of a game. This model captures the dependencies between different stages of the match, providing a detailed probabilistic framework for evaluating player performance. The model's boundary values and probability equations are derived to calculate the winning probabilities for each game.

4.1 Markov Chain Model

A Markov Chain Model is a statistical model that is essentially a stochastic process by which the probability of a transition occurring between two states of a thing is predicted. The state transfer matrix of a Markov chain, the elements of the matrix represent the probability of reaching the next state from the current state. Assuming that the Markov chain has N states, its state transfer matrix can be expressed as:
\begin{equation} P = \left[ {\begin{array}{@{}*{4}{c}@{}} {{p}_{11}}&{{p}_{12}}& \cdots &{{p}_{1N}}\\ {{p}_{21}}&{{p}_{22}}& \cdots &{{p}_{2N}}\\ \vdots & \vdots & \ddots & \vdots \\ {{p}_{N1}}&{{p}_{N2}}& \cdots &{{p}_{NN}} \end{array}} \right] \end{equation}
(1)
The probability of the current state depends only on the previous state and is independent of the historical state. The state transfer probability of a Markov chain is shown in Equation (1):
\begin{equation} {p}_{ij} = \frac{{\sum\limits_{t = 1}^T {I({i}_t = i,{j}_t = j)} }}{{\sum\limits_{t = 1}^T {I({i}_t = i)} }} \end{equation}
(2)
Where I is the indicator function and pij denotes the probability of moving from state i to state j. Next, the smooth distribution vector of this Markov chain is derived, which is used to describe the state distribution of the process in the long run. Its m-step transfer matrix is defined as Pm, which denotes the probability matrix of arriving at a different state after m steps from the current state. Let this Markov chain have a total of N states, then the i-th row and j-th column of its m-step transfer matrix denote the probability of arriving at state j after m steps from state i, as shown in Equation (2):
\begin{equation} {({P}^m)}_{ij} = P({X}_{n + m} = j|{X}_n = i) \end{equation}
(3)
Where Xn denotes the state at the time point n.

4.2 Hierarchical Markov Model

Reinforcing marking the flow of the race when the race point occurs, we extend the Markov chain model and develop a hierarchical Markov chain model.

4.2.1 Modeling games

First, we denote the score as (si, sj, gi, gj, pi, pj). As pi and pj win serve points with probabilities fij, fji, each scoring line (si, sj, gi, gj, pi, pj) in the match goes to one of its two neighboring scoring lines (si, sj, gi, gj, pi+1, pj) and (si, sj, gi, gj, pi, pj+1) with a transition probability that depends on the current server. In a match, the ball is served by either pi or pj. Each game starts at (0, 0) and to win a game, a player must win four or more points by a margin of at least two points. The valid points (pi, pj) for each matchup satisfy pi+ pj >6 | pi - pj <=1, which reduces to (3, 3), (3, 2), or (2, 3). Next, let's take the probability of winning at (3, 3) as an example, where the sender wins the next two points by (fij)2 and the receiver wins the next two points by (1 - fij)2or both players split those points equally, so the probability of the server winning at this point is shown in Equation (3).
\begin{equation} {P}_g(3,3) = \frac{{{{({f}_{ij})}}^2}}{{{{({f}_{ij})}}^2 + {{(1 - {f}_{ij})}}^2}} \end{equation}
(4)
A directed graph of possible scoring order scenarios in the game is shown in Figure 1.
Figure 1:
Figure 1: Game's directed graph
The boundary values are shown in Equation (4):
\begin{equation} {P}_g({p}_i,{p}_j) = \left\{ {\begin{array}{@{}*{1}{c}@{}} {1,{p}_1 = 4,{p}_2 \le 2}\\ {}\\ {0,{p}_2 = 4,{p}_1 \le 2}\\ {}\\ {\frac{{{{({f}_{ij})}}^2}}{{{{({f}_{ij})}}^2 + {{(1 - {f}_{ij})}}^2}},{p}_1 = {p}_2 = 3}\\ {}\\ {{f}_{ij} \times {P}_g({s}_i,{s}_j,{g}_i,{g}_j,{p}_i + 1,{p}_j) + }\\ {(1 - {f}_{ij}){P}_g({s}_i,{s}_j,{g}_i,{g}_j,{p}_i,{p}_j + 1),others} \end{array}} \right. \end{equation}
(5)

4.2.2 Modeling sets

In a set, either pi or pj takes turns serving the ball. Each set starts at (0, 0). To win a set, a player must win six or more sets by at least two sets. If the score is set (6, 6) when both players agree, a special tiebreaker will be played to determine the outcome of the match. Similar to the modeling games, we obtain the boundary values of the set as shown in Equation (5):
\begin{equation} {P}_s({g}_1,{g}_2) = \left\{ {\begin{array}{@{}*{1}{c}@{}} {1,{g}_1 \ge 6,{g}_1 - {g}_2 \ge 2}\\ {}\\ {0,{g}_2 \ge 6,{g}_2 - {g}_1 \ge 2}\\ {}\\ {{P}_t({s}_1,{s}_2),{g}_1 = {g}_2 = {g}_3}\\ {}\\ {{P}_g(0,0)(1 - {P}_s({g}_2,{g}_1 + 1)) + }\\ {(1 - {P}_g(0,0))(1 - {P}_s({g}_2 + 1,{g}_1)),others} \end{array}} \right. \end{equation}
(6)

4.2.3 Modeling a tiebreak game

In a tiebreak game, one player serves the first point, and from there, the two players take turns serving every two points. The game starts at (0, 0) and players have to win by two or more points to win by 7 points. The boundary values are shown in Equation (6).
\begin{equation} \begin{array}{@{}l@{}} {P}_t({p}_1,{p}_2) = \\ \left\{ {\begin{array}{@{}*{1}{c}@{}} {1,{p}_1 \ge 7,{p}_1 - {p}_2 \ge 2}\\ {}\\ {0,{p}_2 \ge 7,{p}_2 - {p}_1 \ge 2}\\ {}\\ {{f}_{ij}{P}_t({g}_1 + 1,{g}_2) + (1 - {f}_{ij}){P}_t({g}_1,{g}_2 + 1),({p}_1 + {p}_2)mod2 = 1}\\ {}\\ {{f}_{ij}(1 - {P}_t({g}_2,{g}_1 + 1)) + (1 - {f}_{ij})(1 - {P}_t({g}_2 + 1,{g}_1)),others} \end{array}} \right. \end{array} \end{equation}
(7)

4.2.4 Modeling a best-of-three match

In a three-game, three-win match, the player who wins the first two games wins. The boundary values of the graph for a three-game, three-win match are shown in Equation (7). In a five-game, three-win match, the boundary values of the graph are s1>=3, s2>=3, because a player must win three games to end the match.
\begin{equation} {P}_m({s}_1,{s}_2) = \left\{ {\begin{array}{@{}*{1}{c}@{}} {1,{s}_1 \ge 2}\\ {0,{s}_2 \ge 2}\\ \begin{array}{@{}l@{}} {P}_s(0,0)({P}_m({s}_1 + 1,{s}_2)) + \\ (1 - {P}_s(0,0))({P}_m({s}_1,{s}_2 + 1)),others \end{array} \end{array}} \right. \end{equation}
(8)

4.2.5 The probability equation for winning

Combining 4.2.1∼4.2.4 above, the winning probability equation for pi, calculated by the recursive algorithm, is shown in Equation (8).
\begin{equation} \begin{array}{@{}l@{}} {P}_m({s}_i,{s}_j,{g}_i,{g}_j,{p}_i,{p}_j) = {f}_{ij} \times {P}_m({s}_i,{s}_j,{g}_i,{g}_j,{p}_i + 1,{p}_j)\\ + (1 - {f}_{ij}){P}_m({s}_i,{s}_j,{g}_i,{g}_j,{p}_i,{p}_j + 1) \end{array} \end{equation}
(9)
We derived the probability of winning for each match. Based on the literature review[7], it is clear that the performance of a player is approximately equal to his probability of winning, and we analyzed and solved for all matches in the dataset. To make the presentation more intuitive and well-formed, we chose the match between Carlos Alcaraz and Novak Djokovic as an example. Visualize the performance of these two players in each of the 5 sets of matches. The performance of Carlos Alcaraz and Novak Djokovic throughout the match is shown in Figure 2. In the visualization, we plotted the average line for each set, where we considered the player with better performance and vice versa.
Figure 2:
Figure 2: Trends in player performance

4.3 Identification of Momentum-Related Factors

To ascertain the most salient factors associated with momentum, a comprehensive analysis was conducted, correlating momentum with eleven distinct indicators. The strength of these associations was quantified and is depicted. Subsequently, the Bootstrap method was employed to statistically validate the existence of a significant relationship between momentum and performance metrics.

4.4 Predicting Momentum Transitions Using LSTM

In the context of tennis match dynamics, particularly when forecasting transitions in momentum, the process is conceptualized as a mapping of information from a singular input to a singular output. To address this, Long Short-Term Memory (LSTM) models were deployed. The LSTM framework is adept at capturing temporal dependencies, making it an ideal candidate for predicting shifts in momentum throughout the course of a match.
For the purpose of model construction and optimization, we curated a dataset comprising all matches featuring Novak Djokovic. A stratified sampling approach was employed, allocating 70% of these matches to form the training set, which was utilized to develop and refine the LSTM model. The remaining 30% of the matches constituted the test set, serving to evaluate the model's predictive efficacy.
In the model construction phase, momentum was designated as the primary input variable for the LSTM model. This choice was motivated by the hypothesis that momentum, as a dynamic and influential factor in match progression, holds predictive power over match outcomes.

4.5 Strategic Implications for Coaches

Our analysis identified a significant correlation between Unforced Error and Momentum, indicating that Unforced Error is a pivotal determinant of Momentum in a match. An elevated rate of Unforced Errors is typically indicative of a decline in Momentum. Additionally, factors such as Winner, Rally Count, Distance Run, game_victor, and speed_mph, as well as Break Point Won, demonstrated strong relationships with Momentum. These findings underscore the significance of controlling errors, enhancing winning points, sustaining match intensity, performing effectively at critical junctures, and maintaining high stroke velocities in the prediction and management of match Momentum.
Figure 3:
Figure 3: Result comparation between Real Momentum and Predict Momentum
In addition, we counted the number of times the serve side won the match point set for each game. The vast majority of players serving in match point sets performed above the average performance score for this game. Carlos Alcaraz was better in the 2nd and 3rd set and Novak Djokovic was better in the 1st, 4th and 5th set. During the match point set, it was calculated that Carlos Alcaraz and Novak Djokovic had 77.27% and 79.17% probability of winning the serve and 20.83% and 22.73% probability of winning the receive respectively. The results show that the probability of serve winning in match point sets is significantly higher.
While other variables like set_victor, Ace, Break Point Missed, and Double Fault also exhibit associations with Momentum, their impact is less pronounced compared to the aforementioned factors. Coaches can leverage these significantly correlated metrics to adjust game strategies in real time. For instance, an increase in the opponent's Unforced Errors presents an opportune moment to intensify offensive play. Conversely, when a player experiences a surge in Unforced Errors, it is crucial to stabilize the situation. By doing so, coaches can assist players in better managing the game's tempo and making strategic adjustments at pivotal moments, thereby leveraging or reversing the Momentum of the match.

5 RESULTS and DISCUSSIONs

5.1 Model Evaluation and Generalization

Incorporating image analysis and statistical measures such as R-squared (R2), the performance metrics of our model are detailed in Table 2. The model exhibits robust generalization capabilities with respect to the playing surface and gender of tennis players. However, the predictive accuracy of the model warrants enhancement when applied to different types of balls and game systems.
Table 2:
PlayerR2RMSEPlayerR2RMSE
HC0.95060.5642EF10.52362.1265
CC0.95110.5267EF20.50192.5611
GC0.94560.5549EF30.49253.1021
FT10.92190.6306TT10.47692.4263
FT20.92860.5872TT20.48722.5106
FT30.93010.6103TT30.45362.7126
Table 2: Error Analysis

5.2 Model Limitations

Upon reviewing the existing literature and synthesizing our findings, we conducted an analysis of the model's limitations in various sports contexts:
Tournaments are structured to identify the ultimate victor through concentrated competition within a brief timeframe, necessitating athletes to possess formidable mental fortitude and adaptable tactical acumen. In contrast, championship events span a longer duration, with athletes accruing points throughout the season and the final standings determined by the aggregate points at its conclusion. These events demand a sustained level of performance stability from the athletes. Consequently, our model encounters certain constraints in forecasting the trajectory of momentum changes among table tennis tournament participants.
Soccer championships represent a collective sport event where the influence of an individual athlete on the game's progression is considerably less pronounced than that of an individual competitor. Given the substantial impact of scoring events in soccer, they may introduce considerable variance into the prediction outcomes.

6 Conclusions

6.1 Strengthen

The Hierarchical Markov Model offers several strengths in modeling tennis matches:
It captures both short-term dependencies, such as game-to-game transitions, and long-term dependencies, such as set-to-set progressions, at different hierarchical levels, enhancing the accuracy of probabilistic match outcome predictions.
HMM effectively models state transitions within a game, such as shifts from defense to offense or possession to turnover, providing insights into the game's fluidity.
The model's hierarchical structure adheres to the rules of tennis, allowing for precise calculations and predictions of win probabilities at various stages of the game.

6.2 Weaknesses

The predictive accuracy of HMM is contingent upon the validity of the assumptions regarding match progression and the state transition matrix. Erroneous assumptions can compromise the model's precision.

6.3 Extensions

The HMM's hierarchical framework has applications beyond sports. In the field of medicine, it can predict the progression of conditions stage by stage, aiding doctors in understanding patient conditions and formulating treatment strategies.The model is equally applicable to other stage-structured tournaments, such as soccer and basketball, offering real-time game analysis and win rate predictions.

6.4 LSTM Capabilities

LSTM models, while not explicitly discussed in the original text, complement HMM by:
Handling multiple input data types, including match statistics and player performance data.
Learning long-term dependencies and adapting to variable time steps, ensuring flexibility with time series data without loss of critical information.
This study presents an innovative application of HMM for evaluating tennis player performance. The model's ability to analyze game flow and state transitions provides a comprehensive understanding of match dynamics, revealing new insights into strategic and performance patterns.
Future research should integrate Markov model outcomes with momentum metrics for a deeper analysis of athlete performance. Momentum metrics can offer coaches and athletes additional insights into an athlete's status and identify pivotal game moments. Such an integrated analysis could inform the development of more effective training strategies, enhancing athlete performance in competitions and contributing to the advancement of sports performance analysis.

Acknowledgments

This research was funded by Wuhan City Polytechnic Research Program for Master's and Doctor's Projects in 2023, funding reference number: 2023whcvcB07; This research has received substantial support from: Research Project of Hubei Provincial China Vocational Education Association in 2024, funding reference number: HBZJ2024317, and Hubei Provincial Education Department Research Project: 110KW Substation Transformer Oil Leakage Abnormal Condition Monitoring System Based on Deep Learning, number B2021362.

Footnotes

1
http://www.comapmath.com/MCMICM/Wimbledon_featured_matches.csv
2
https://www.immchallenge.org/mcm/data_dictionary.csv

References

[1]
Wunderlich F, Memmert D. Forecasting the outcomes of sports events: A review[J]. European journal of sport science, 2021, 21(7): 944-957.
[2]
Rico-Gonzalez M, Pino-Ortega J, Praca G M, et al. Practical applications for designing soccer'training tasks from multivariate data analysis: a systematic review emphasizing tactical training[J]. Perceptual and Motor Skills, 2022, 129(3): 892-931.
[3]
Huang M L, Lin Y J. Regression tree model for predicting game scores for the golden state warriors in the national basketball association[J]. Symmetry, 2020, 12(5): 835.
[4]
Rajšp A, Fister Jr I. A systematic literature review of intelligent data analysis methods for smart sport training[J]. Applied Sciences, 2020, 10(9): 3013.
[5]
Patel D, Shah D, Shah M. The intertwine of brain and body: a quantitative analysis on how big data influences the system of sports[J]. Annals of Data Science, 2020, 7(1): 1-16.
[6]
Chmait N, Westerbeek H. Artificial intelligence and machine learning in sport research: An introduction for non-data scientists[J]. Frontiers in sports and active living, 2021, 3: 363.
[7]
Hills S P, Barrett S, Busby M, et al. Profiling the post-match top-up conditioning practices of professional soccer substitutes: An analysis of contextual influences[J]. The Journal of Strength & Conditioning Research, 2020, 34(10): 2805-2814.
[8]
8.Chan, T. C. Y., Fearing, D. S., Fernandes, C., & Kovalchik, S. (2021). "A Markov Process Approach to Untangling Intention versus Execution in Tennis." University of Toronto (Canada), ProQuest Dissertation & Theses, 28767942.
[9]
9.Sim, M. K., & Choi, D. G. (2019). "The Winning Probability of a Game and the Importance of Points in Tennis Matches." Journal of Sports Science and Medicine, 19(3), 361–372.
[10]
Gollub, J. (2021). "Forecasting serve performance in professional tennis matches." Journal of Sports Analytics, 7(4), 223-233.
[11]
11.Yu, S., Sun, J., & Hu, Y. (2022). "Visualisation of Athlete Performance Based on Markov Chains and EWMs." Journal of Sports Science, 29(6), 1-12.
[12]
12.Tea, P., & Swartz, T.B. (2023). "The analysis of serve decisions in tennis using Bayesian hierarchical models." Annals of Operations Research, 325, 633–648.
[13]
13.Fernandes, C. J. (2021). "A Markov Process Approach to Untangling Intention versus Execution in Tennis." University of Toronto (Canada), ProQuest Dissertation & Theses, 28767942.
[14]
14. Meier, P., Flepp, R., Ruedisser, M., & Franck, E. (2020). Separating psychological momentum from strategic momentum: Evidence from men's professional tennis. Journal of Economic Psychology, 102269.
[15]
Tod D. Sport psychology: The basics[M]. Routledge, 2022.
[16]
Morgulev, E., Azar, O. H., & Bar-Eli, M. (2020). Searching for momentum in NBA triplets of free throws. Journal of Sports Sciences, 38(4), 390–398.
[17]
Dietl H, Nesseler C. Momentum in tennis: Controlling the match[J]. UZH Business Working Paper Series, 2017 (365).
[18]
Iso-Ahola S E, Dotson C O. Psychological momentum—A key to continued success[J]. Frontiers in psychology, 2016, 7: 214368.
[19]
S. Han, Alia Shayahaxi, and C. Yu, “Research on Predicting Momentum Changes in Tennis Matches Based on Markov and Random Forest Algorithms”, HSET, vol. 94, pp. 586–595, Apr. 2024.
[20]
Koski J V, Landig A J, Pályi A, et al. Floquet spectroscopy of a strongly driven quantum dot charge qubit with a microwave resonator[J]. Physical Review Letters, 2018, 121(4): 043603.
[21]
Zhang Q, Zhang X, Hu H, et al. Sports match prediction model for training and exercise using attention-based LSTM network[J]. Digital Communications and Networks, 2022, 8(4): 508-515.
[22]
Bunker R, Susnjak T. The application of machine learning techniques for predicting match results in team sport: A review[J]. Journal of Artificial Intelligence Research, 2022, 73: 1285-1322.
[23]
Alzubaidi L, Zhang J, Humaidi A J, et al. Review of deep learning: concepts, CNN architectures, challenges, applications, future directions[J]. Journal of big Data, 2021, 8: 1-74.

Recommendations

Comments

Information & Contributors

Information

Published In

cover image ACM Other conferences
ICIIP '24: Proceedings of the 2024 9th International Conference on Intelligent Information Processing
November 2024
419 pages
ISBN:9798400718076
DOI:10.1145/3696952

Publisher

Association for Computing Machinery

New York, NY, United States

Publication History

Published: 21 November 2024

Check for updates

Author Tags

  1. Data-Driven Strategy
  2. Hierarchical Markov Models
  3. Match Dynamics Performance Evaluation
  4. Sports Analysis

Qualifiers

  • Research-article

Conference

ICIIP 2024

Acceptance Rates

Overall Acceptance Rate 87 of 367 submissions, 24%

Contributors

Other Metrics

Bibliometrics & Citations

Bibliometrics

Article Metrics

  • 0
    Total Citations
  • 247
    Total Downloads
  • Downloads (Last 12 months)247
  • Downloads (Last 6 weeks)93
Reflects downloads up to 17 Feb 2025

Other Metrics

Citations

View Options

View options

PDF

View or Download as a PDF file.

PDF

eReader

View online with eReader.

eReader

HTML Format

View this article in HTML Format.

HTML Format

Login options

Figures

Tables

Media

Share

Share

Share this Publication link

Share on social media