Прогнозування результатів тенісних матчів і аналіз фінансових вигод
Tennis is one of the most popular sports in the world, attracting considerable attention from casual fans and professional analysts. The application of machine learning methods enables the accurate prediction of match results, opening up opportunities for profit through betting on likely winners. Th...
Saved in:
| Date: | 2025 |
|---|---|
| Main Authors: | , |
| Format: | Article |
| Language: | English |
| Published: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2025
|
| Subjects: | |
| Online Access: | https://journal.iasa.kpi.ua/article/view/343065 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | System research and information technologies |
| Download file: |  |
Institution
System research and information technologies| _version_ | 1866303049660628992 |
|---|---|
| author | Shum, Kyryl Kuznietsova, Nataliia |
| author_facet | Shum, Kyryl Kuznietsova, Nataliia |
| author_sort | Shum, Kyryl |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2025-11-09T00:01:30Z |
| description | Tennis is one of the most popular sports in the world, attracting considerable attention from casual fans and professional analysts. The application of machine learning methods enables the accurate prediction of match results, opening up opportunities for profit through betting on likely winners. This study evaluates the financial benefits of predicting tennis match outcomes by identifying an effective sports betting strategy. The study examines various machine learning methods and auxiliary algorithms, comparing them to select the best betting strategy for maximizing the user’s potential profit. In the paper, the method and algorithm for determining effective sports betting strategies were developed. This algorithm and method were tested on tennis game datasets (for both women and men), and the best tennis betting strategy was identified. As part of the study, a software product has been developed to predict the outcomes of tennis matches. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2025.3.07 |
| first_indexed | 2025-11-09T02:11:03Z |
| format | Article |
| fulltext |
К. Shum, N. Kuznietsova, 2025
Системні дослідження та інформаційні технології, 2025, № 3 87
UDC 004.8:336:796.3:519.83
DOI: 10.20535/SRIT.2308-8893.2025.3.07
ANALYSIS AND FORECASTING OF THE FINANCIAL BENEFIT
FOR THE TENNIS MATCH OUTCOMES BY MACHINE
LEARNING METHODS
К. SHUM, N. KUZNIETSOVA
Abstract. Tennis is one of the most popular sports in the world, attracting consider-
able attention from casual fans and professional analysts. The application of ma-
chine learning methods enables the accurate prediction of match results, opening up
opportunities for profit through betting on likely winners. This study evaluates the
financial benefits of predicting tennis match outcomes by identifying an effective
sports betting strategy. The study examines various machine learning methods and
auxiliary algorithms, comparing them to select the best betting strategy for maximiz-
ing the user’s potential profit. In the paper, the method and algorithm for determin-
ing effective sports betting strategies were developed. This algorithm and method
were tested on tennis game datasets (for both women and men), and the best tennis
betting strategy was identified. As part of the study, a software product has been de-
veloped to predict the outcomes of tennis matches.
Keywords: forecasting, machine learning, betting strategies, financial benefit.
INTRODUCTION
Tennis is a dynamic and unpredictable game, combining many factors that influ-
ence the course of events during matches: players’ physical conditions, psycho-
logical state, chosen tactics, anthropometry, weather conditions, and more. Each
of these aspects can be decisive in achieving the desired outcome. Thanks to this
versatility, tennis ranks among the most popular sports globally, captivating a
broad audience of fans, from casual spectators who enjoy the thrill of the game to
professional sports analysts who study the game from a scientific perspective.
Match outcome prediction holds a special place among the various aspects of
sports interest. As for the standard fan, a match result is typically a topic of dis-
cussion and emotional enjoyment. However, the prediction is practical for ana-
lysts and professional bettors who place wagers on sports events [1]. Knowing the
likelihood of a player’s victory not only allows for more informed betting to se-
cure financial gain but also aids in developing strategies for long-term success. In
this context, the betting process goes beyond simple gambling for many profes-
sional participants in the sports betting market. Predicting tennis match outcomes
becomes a critical tool for making informed decisions, assessing risks, and evalu-
ating potential benefits, ultimately supporting the financial growth of the bettor. It
also gives us the possibility to solve such tasks as understanding behavior and
forecasting the gamer’s outflow [2]. The players who win are motivated to stay
longer in the game while they understand the game’s process and can also plan
their own strategy and evaluate their financial benefits.
К. Shum, N. Kuznietsova
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 88
PROBLEM STATEMENT
This research was conducted to deeply understand the betting and gambling proc-
esses by applying modern techniques and approaches. It was first decided to try
machine learning algorithms for evaluating and forecasting games’ outcomes and
for finding hidden dependencies. Then, based on these models, we can find the
most important variables that could be interpreted as some key factors for win-
ning on some side. It means that we can also take into account some preliminary
information before making a bet. Next, the strategy of effective betting should be
defined. For this reason, we will develop the algorithm for defining the most ef-
fective strategy that can be used by gamblers to maximize their profit as a result
of the sports betting process.
MACHINE LEARNING METHODS AND AUXILIARY ALGORITHMS
FOR THE GAME OUTCOMES PREDICTION
Machine learning methods
Machine learning is currently a powerful tool for solving various tasks across dif-
ferent fields of human activity. The significant potential and efficiency of ma-
chine learning methods and algorithms make this technology crucial in areas
where traditional approaches may fall short. Predicting the outcomes of sports
events is no exception. In this study, predictive models have been developed to
predict the results of men’s (hereafter, M) and women’s (hereafter, W) tennis
matches. These models are based on logistic regression (M and W), multilayer
perceptron (W), random forest (M), and extreme gradient boosting (M).
Logistic regression is a method that models the relationship between a cate-
gorical target variable and a set of independent predictor variables. Although lo-
gistic regression is a classification algorithm, it is based on a linear regression
model [3; 4]. To produce categorical outcomes, it transforms the continuous out-
put of linear regression into a range between 0 and 1 (interpreted as the probabil-
ity of belonging to a specific class) using the logistic function, also known as the
sigmoid function [3], which can be described by the following formula:
,
1
1
)(
ze
z
where nnxxxz 22110 is a linear combination of independent
variables ix and their coefficients nii , 1, , n is a number of predictor vari-
ables and 0 is the intercept term.
Then the probability of an object’s belonging to a specific class can be repre-
sented as:
)()1( zyP ; )(1)0( zyP .
Random forest is an ensemble machine learning method that combines the
predictions of multiple decision trees to improve the model’s accuracy and stabil-
ity. The trees are constructed on random subsets of data from the training set, and
random subsets of features are used to reduce the correlation between the trees.
The final prediction ŷ is determined by majority voting, making the method ro-
bust to overfitting and effective for various tasks [5], and is determined by the
following formula:
Analysis and forecasting of the financial benefit for the tennis match outcomes by…
Системні дослідження та інформаційні технології, 2025, № 3 89
, 1, )}({ˆ nixhModey i ,
where )(xhi is a prediction of the i -th tree, аnd n is a number of trees in the forest.
Extreme Gradient Boosting (XGBoost) is an ensemble machine learning
method that implements gradient boosting with decision trees. XGBoost uses an
iterative approach, where the key idea is to build an ensemble of decision trees,
with each subsequent tree sequentially correcting the errors of the previous ones,
thereby improving the model’s overall accuracy [6]. If the prediction for the i -th
sample after 1k iterations is represented as )1(ˆ k
iy , then at the k -th iteration,
the prediction value will be updated using the following formula:
,)(ˆˆ )1()
ik
k
i
k
i xhyy
where is the learningn rate, which determines how strongly each tree influences
the final prediction.
A multilayer perceptron (MLP) is a type of artificial neural network consist-
ing of several layers: an input layer, one or more hidden layers, and an output lay-
er. Each layer contains neurons that take a weighted sum of input data from the
previous layer, apply an activation function to it, and pass the result to the next
layer. Weighted sum is counted using the following formula:
, ) ()1()(
1
)( l
j
l
i
l
ij
n
i
l
j bawz
where )(l
jz is the activation of neuron j in layer l ; )(l
ijw is the weight connecting
neuron i in the previous layer to the neuron j ; )1( l
ia is the activation of neuron
i in the previous layer; ) (l
jb is the bias of neuron j .
The training process is repeated over several iterations until the model con-
verges to an optimal solution. Due to the architecture, MLP can model complex
nonlinear relationships between data [7].
Auxiliary algorithms
The time discounting method is an approach that assigns greater significance to
newer data and less to older data [8]. The idea is to apply weights relative to the
time between events. In the study, this method is used to predict player statistics
in the men’s division, as the prediction of match performance is based on the val-
ues of relevant statistical variables. Weights are applied using an exponential
function )(tW :
), (min)( fftW t ,
where t represents the time in months between the scheduled match and a previ-
ously played match, and f is the discount factor, which can range from 0 to 1.
The discount factor determines the extent of time discounting and is set by the re-
searcher. The smaller the value of f , the lower the weight given to older matches.
The data filtering algorithm is the process of selecting subsets from a large
dataset based on specified criteria. Initially, the selection conditions are defined
(for example, these may be the values of variables), after which the corresponding
samples are formed, which allows the efficient extraction of the most relevant
data for further analysis and use.
К. Shum, N. Kuznietsova
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 90
BUILDING THE MACHINE LEARNING MODELS
For this study, we decided to use real data and develop our models for the predic-
tion of tennis match outcomes using Python, along with relevant machine learning
and data processing libraries (Sklearn, Pandas, Numpy, etc.). For this reason, we
used two different datasets for men’s and women’s games. The dataset from user
JeffSackmann’s GitHub repository [9] was used to predict men’s matches, and the
dataset from the Tennis-data website [10] was utilized for women’s matches. In
both cases, the records began at the start of 2010, with a total of 153.959 and
37.731 games recorded, respectively.
For both datasets, initial preprocessing was carried out: the properties and
specifics of each variable were analyzed, missing values were handled, irrelevant
records and variables were removed, and the data was transformed into a format
suitable for future models. Each dataset was duplicated, and the corresponding
player columns were swapped to balance the number of positive and negative
classes (1 for the first player’s victory, 0 for loss). As a result, the final training
datasets contained 209.116 and 47.816 records for men and women, respectively.
The most important features for prediction by using statistical methods were
selected, such as:
For men: twenty significant predictor variables were selected, including
tournament seeding numbers, differences in height, ranking, ranking points, as
well as percentage differences in various statistical indicators (e.g., first serve
percentage, percentage of points won on return, etc.).
For women: five significant predictor variables were selected, including
differences in ranking points, age, and differences in the win odds set by the Pinnacle
bookmaker and between maximum and average odds from other bookmakers.
The next stage involves constructing machine learning models using the
methods mentioned earlier. To determine their best parameters, the grid search
algorithm was applied. This algorithm selects the combination of the most effec-
tive features from a given set that ensures the highest model performance. The
results obtained are presented in Tables 1–5.
T a b l e 1 . Parameters of the logistic regression model (women)
Parameter Description Value
test_size The proportion of the dataset that is used for testing the model 0.1
solver The method for determining the optimal model weights
that minimize the loss function liblinear
fit_intercept The presence of a bias term in the model equation False
C Regularization strength 4.25
penalty The type of regularization used to control the model’s overfitting L2
T a b l e 2 . Parameters of the multilayer perceptron model (women)
Parameter Description Value
test_size The proportion of the dataset that is used for testing the model 0.1
solver The method for determining the optimal model weights
that minimize the loss function lbfgs
activation Activation function of the hidden layer relu
alpha L2-regularization strength 0.005
hidden_layer_ sizes Number of neurons in the hidden layers (100,)
learning_rate Learning rate for weight updates constant
Analysis and forecasting of the financial benefit for the tennis match outcomes by…
Системні дослідження та інформаційні технології, 2025, № 3 91
T a b l e 3 . Parameters of the logistic regression model (men)
Parameter Description Value
test_size The proportion of the dataset that is used for testing the model 0.1
solver The method for determining the optimal model weights that
minimize the loss function newton-cg
fit_intercept The presence of a bias term in the model equation True
C Regularization strength 1
penalty The type of regularization used to control the model’s overfitting None
T a b l e 4 . Parameters of the random forest model (men)
Parameter Description Value
test_size The proportion of the dataset that is used for testing the model 0.15
n_estimators The number of trees in the forest 100
criterion The function to measure the quality of a split log_loss
max_features The number of features to consider when looking for the best split None
min_samples_leaf The minimum number of samples required to be at a leaf node 2
T a b l e 5 . Parameters of the XGBoost model (men)
Parameter Description Value
test_size The proportion of the dataset that is used for testing the model 0.15
n_estimators The number of trees (iterations) of the model 100
learning_rate Learning rate 0.05
max_depth The maximum depth of each tree 6
reg_alpha L1-regularization parameter 0.5
reg_lambda L2-regularization parameter 1.5
After building the models with the specified parameters, their performance
was evaluated on the training and validation datasets using standard classification
quality metrics. The results are presented in Tables 6–10.
T a b l e 6 . Evaluation of the logistic regression model (women)
Quality metric
Sample
Accuracy Precision Recall F1 Score Roc Auc Loss
Training 0.68966 0.68897 0.70124 0.69496 0.76102 0.57987
Validation 0.68341 0.68355 0.68202 0.68277 0.75474 0.58704
T a b l e 7 . Evaluation of the multilayer perceptron model (women)
Quality metric
Sample
Accuracy Precision Recall F1 Score Roc Auc Loss
Training 0.69175 0.68903 0.70954 0.69894 0.7599 0.5803
Validation 0.68290 0.68239 0.68323 0.68274 0.7535 0.58745
К. Shum, N. Kuznietsova
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 92
T a b l e 8 . Evaluation of the logistic regression model (men)
Quality metric
Sample
Accuracy Precision Recall F1 Score Roc Auc Loss
Training 0.98278 0.98292 0.98253 0.98272 0.99790 0.05599
Validation 0.98145 0.98146 0.98144 0.98145 0.99789 0.05618
T a b l e 9 . Evaluation of the random forest model (men)
Quality metric
Sample
Accuracy Precision Recall F1 Score Roc Auc Loss
Training 0.98345 0.98345 0.98376 0.98381 0.99763 0.07793
Validation 0.98363 0.98362 0.98366 0.98349 0.99789 0.07484
T a b l e 1 0 . Evaluation of the XGBoost model (men)
Quality metric
Sample
Accuracy Precision Recall F1 Score Roc Auc Loss
Training 0.98396 0.98501 0.98293 0.98396 0.99834 0.04733
Validation 0.98359 0.98267 0.98455 0.98361 0.99842 0.04601
After analyzing the results obtained, it can be concluded that the models pre-
dicting women’s matches perform at an acceptable level but are slightly worse than
those predicting men’s matches. The models for men’s tennis show excellent values
across all metrics. However, it is important to note that their performance may decline
due to the necessity of applying the time discounting method to predict statistics for
future matches. None of the developed models exhibit signs of overfitting, as the
quality metrics for both the training and validation datasets are very close.
To facilitate the process of predicting matches and to provide a straightfor-
ward interpretation of the results, a web interface was developed to allow users to
interact with the developed models easily, input the necessary data for predictions
via the keyboard, and modify it if needed. Separate prediction pages for men and
women were implemented, with their interfaces shown in Figs. 1 and 2. Addition-
ally, a database containing historical match records for men was created.
Fig. 1. Prediction page for men’s match outcomes
Analysis and forecasting of the financial benefit for the tennis match outcomes by…
Системні дослідження та інформаційні технології, 2025, № 3 93
ALGORITHM FOR DETERMINING THE BEST BETTING STRATEGIES
To determine an effective betting strategy, a method based on the ROI (return on
investment) metric as the prior indicator of a bettor’s success and, accordingly,
the target metric of the built predictive model’s effectiveness was developed and
applied. An important condition is that each bet must be evenly distributed with
an identical amount.
Let 0S represent the bettor’s (player’s) initial capital.
,0 tsS
where s is the amount of a single bet (always the same), and t is the bettor’s tol-
erance for losses, i.e., the number of consecutive bets he is willing to lose before
ceasing to follow the strategy. The tolerance is determined by the bettor and can
be adjusted during the betting process.
Then, iS is the player’s current capital after the i -th bet has been placed.
,1 scoefsSS iiii
where icoef is the coefficient of the i -th bet, ni , 1 , and n is the total number of bets,
while i is the indicator of the success of the i -th bet, which is determined as follows:
otherwise, 0 won bettheif , 1{i .
Let iP be the player’s profit after calculating the i -th bet:
0SSP ii .
Let iROI be the percentage of winnings from each bet, calculated after the
i -th bet, averaged over the distance:
ni
is
P
ROI i
i , 1,
100
.
Then, the betting strategy is considered effective if the following conditions are met:
1. nisSi , 1, . This condition means that the player’s current capital
must always be greater than the amount of one bet to be able to place it.
Fig. 2. Prediction page for women’s match outcomes
К. Shum, N. Kuznietsova
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 94
2. nhiROIi , при 0 . Here, n is the total number of bets placed, and
h is the minimum number of bets determined as the calculation threshold for
profit, which can be adjusted by the player. This condition means that after the h -
th bet, the return on investment (ROI) must always be greater than 0, demonstrat-
ing the strategy’s stability and profitability over the long term.
3. ,)(1 1
1
0
tROIROI jkjk
t
j
where 1, 1 tnk for ,0, 1 0 ROItn and )(1 is an indicator defined as
follows:
otherwise, 0 if, 1{)(1 1 jkjk ROIROI .
This condition means that the player is willing to tolerate no more than t
consecutive lost bets (tolerance for loss).
The algorithm for determining an effective strategy using the described
method is presented in the form of a flowchart in Fig. 3.
Fig. 3. Flowchart of the algorithm for determining the effectiveness of a betting strategy
Analysis and forecasting of the financial benefit for the tennis match outcomes by…
Системні дослідження та інформаційні технології, 2025, № 3 95
DETERMINING THE BEST BETTING STRATEGIES BY DEVELOPED
ALGORITHM
New datasets of predicted matches were created using the models and algorithms
previously developed to determine potentially successful betting strategies. A
sample of 239 predicted matches for women and 346 for men was compiled. The
prediction of players’ statistics for men was performed in two algorithm varia-
tions: for all court surfaces and only for a selected surface. As a result, six predic-
tions were made for each men’s match.
A filtering process was applied to the obtained data to exclude specific cate-
gories of games, thereby increasing the scope for identifying the most suitable
conditions for profitability. For women, filters were considered for the minimum
probability of a player’s victory and the minimum odds. For men, filters included
the current form (i.e., the number of matches played in the last 60 days) and the
minimum odds. Tables 11–12 present the number of successful (profitable) strat-
egies for each model.
T a b l e 1 1 . Successful strategies (women)
Model Number of profitable
strategies
Total number
of strategies
Percentage of
profitable strategies
Logistic regression 44 99 44
Multilayer perceptron 34 99 34
T a b l e 1 2 . Successful strategies (men)
Model Number of
profitable strategies
Total number
of strategies
Percentage of
profitable strategies
Logistic regression (all surfaces) 24 144 17
Random forest (all surfaces) 33 144 23
XGBoost (all surfaces) 0 144 0
Logistic regression (selected surface) 18 99 18
Random forest (selected surface) 0 99 0
XGBoost (selected surface) 7 99 7
The tables reveal that the XGBoost model with the algorithm for predicting
players’ statistics across all surfaces and the random forest model with the algo-
rithm for predicting players’ statistics on a selected surface did not show any prof-
itable betting strategies.
Tables 13–14 present the most successful and effective strategies for each
model with the parameters .10 ,5, 100 hts
T a b l e 1 3 . Most successful effective strategies for the women’s division
Model
Minimum
probability
threshold
Minimum
coefficient
threshold
Prediction ratio
(correct predictions
/ total predictions)
Percentage
of correct
predictions
Increase
in initial
capital (%)
ROI
(%)
Logistic
regression
0.65 1.35 22/23 96 154.8 33.65
Multilayer
perceptron
0.65 1.35 27/28 96 201.8 36.04
К. Shum, N. Kuznietsova
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 96
T a b l e 1 4 . Most successful effective strategies for the men’s division
Model
Minimum
number of
matches played
threshold
Minimum
coefficient
threshold
Prediction ratio
(correct predic-
tions / total
predictions)
Percentage
of correct
predictions
Increase
in initial
capital
(%)
ROI
(%)
Logistic
regression
(all surfaces)
12 1.5 14/24 57 57.4 11.96
Random forest
(all surfaces) 12 1.25 19/30 63 158.6 26.43
Logistic
regression
(selected surface)
8 1.6 12/19 63 82 21.58
XGBoost (se-
lected surface) 8 1.55 14/23 61 104.2 22.48
Based on the results, it can be concluded that the two best strategies for ob-
taining financial gains from betting on tennis match outcomes are:
For women: the multilayer perceptron model, as its strategy has a higher
percentage increase in initial capital and ROI.
For men: the random forest model with the algorithm for predicting
players’ statistics across all types of courts, as it has the highest ROI and percent-
age increase in initial capital.
To visualize the change in ROI from betting according to the most success-
ful effective strategies, we generated the graphs shown in Fig. 4 and 5, illustrating
the effectiveness of the chosen strategies and models for women’s and men’s ten-
nis matches, respectively.
Fig. 4. Change in ROI for the most successful effective strategy based on the multilayer
perceptron model for women’s tennis matches
Fig. 5. Change in ROI for the most successful effective strategy based on the random
forest model with statistical prediction algorithm across all types of courts for men’s tennis
matches
Analysis and forecasting of the financial benefit for the tennis match outcomes by…
Системні дослідження та інформаційні технології, 2025, № 3 97
CONCLUSIONS
The first part of the conducted study was dedicated to the implementation of vari-
ous machine learning methods and auxiliary algorithms for predicting the out-
comes of tennis matches. The second part aimed to determine the best strategies
for obtaining financial benefit from sports betting. For this work the real data both
for men’s and women’s tennis matches were selected, processed and analyzed.
Five machine learning models were developed based on logistic regression, multi-
layer perceptron, random forest, and extreme gradient boosting methods. Men’s
tennis results forecasting is based on players’ statistics as predictor variables.
Therefore, an algorithm that uses the time discounting method was applied, ena-
bling the statistics forecasting for future matches based on the player’s historical
games. Forecasting of outcomes were made on new datasets to determine the best
betting strategies. Based on the results obtained, using a filtering algorithm and
the developed method for assessing strategy effectiveness, the most successful
and effective betting strategies were identified for use in sports betting to maxi-
mize user profits.
A web interface was created to facilitate the use of the developed models and
provide a clear interpretation of the obtained results. This interface allows users to
easily manipulate input data for prediction by entering it via the keyboard or, if
necessary, modifying it. In future research, we will focus on studying and using
background information received from the key variables as well as modifying and
proposing more different strategies for the players based on their attitude and risk
tolerance.
REFERENCES
1. Gianluca Di Censo, Paul Delfabbro, Daniel King, “Examining the Role of Sports
Betting Marketing in Youth Problem Gambling,” Journal of Gambling Studies,
vol.40, pp. 2005–2025, 2024. doi: http://dx.doi.org/10.1007/s10899-024-10347-x
2. N. Kuznietsova, P. Bidyuk, “Forecasting of Financial Risk Users’ Outflow,” IEEE
First International Conference on System Analysis & Intelligent Computing (SAIC),
Kyiv: 08–12 October, 2018, pp. 250–255. Available: https://ieeexplore.ieee.org/ab-
stract/document/8516782
3. A.T. Fernandes, D.B. Figueiredo Filho, E.C. da Rocha, W. da S. Nascimento, “Read
this paper if you want to learn logistic regression,” Revista de Sociologia e Política,
vol. 28, no. 74, 2020. doi: https://doi.org/10.1590/1678-987320287406en
4. N.V. Kuznietsova, Z.S. Chernysh, “Regression models application for analysis and
forecasting of the financial activity quality indicators of the company,” System Re-
search and Information Technologies, no. 2, pp. 67–81, 2020. doi:
https://doi.org/10.20535/SRIT.2308-8893.2020.2.05
5. M. Brown, Applied Predictive Analytics. 2023. [Online]. Available: https://
core.ac.uk/download/568285513.pdf
6. C. Bentéjac, A. Csörgő, G. Martínez-Muñoz, “A comparative analysis of gradient
boosting algorithms,” Artificial Intelligence Review, vol. 54, no. 3, Aug. 2020. doi:
https://doi.org/10.1007/s10462-020-09896-5
7. K.Y. Chan et al., “Deep neural networks in the cloud: Review, applications, chal-
lenges and research directions,” Neurocomputing, vol. 545, p. 126327, Aug. 2023.
doi: https://doi.org/10.1016/j.neucom.2023.126327
К. Shum, N. Kuznietsova
ISSN 1681–6048 System Research & Information Technologies, 2025, № 3 98
8. M. Sipko, W. Knottenbelt, MEng Computing -Final year project Machine Learning
for the Prediction of Professional Tennis Matches. 2015. Available: https://
www.doc.ic.ac.uk/teaching/distinguished-projects/2015/m.sipko.pdf
9. “GitHub - JeffSackmann/tennis_atp: ATP Tennis Rankings, Results, and Stats,”
GitHub. Available: https://github.com/JeffSackmann/tennis_atp
10. “Tennis Betting | Tennis Results | Tennis Odds,” Tennis-data.co.uk, 2019. Available:
http://www.tennis-data.co.uk/alldata.php
Received 24.12.2024
INFORMATION ON THE ARTICLE
Kyryl I. Shum, ORCID: 0009-0009-7503-3249, Educational and Research Institute for
Applied System Analysis of the National Technical University of Ukraine “Igor Sikorsky
Kyiv Polytechnic Institute”, Ukraine, e-mail: shumkirillid@gmail.com
Nataliia V. Kuznietsova, ORCID: 0000-0002-1662-1974, Educational and Research
Institute for Applied System Analysis of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: natalia-kpi@ukr.net
ПРОГНОЗУВАННЯ РЕЗУЛЬТАТІВ ТЕНІСНИХ МАТЧІВ І АНАЛІЗ
ФІНАНСОВИХ ВИГОД / К.І. Шум, Н.В. Кузнєцова
Анотація. Реалізовано програмний продукт, який дозволяє прогнозувати ре-
зультати тенісних матчів, розроблено метод визначення ефективних стратегій
спортивних ставок. Теніс є одним із найпопулярніших видів спорту у світі,
який привертає значну увагу як звичайних уболівальників, так і професійних
аналітиків. Використання методів машинного навчання дає змогу ефективно
прогнозувати результати матчів, що відкриває можливості для отримання при-
бутку від ставок на ймовірних переможців. Мета дослідження – оцінювання
фінансової вигоди від прогнозування результатів тенісних ігор через пошук
ефективної стратегії спортивних ставок. Розглянуто різні методи машинного
навчання і допоміжні алгоритми та виконується їх порівняння з метою вибору
найкращої стратегії укладання ставок задля максимізації потенційного прибу-
тку користувача. Об’єкт дослідження – прогнозування результативності теніс-
них матчів. Предмет дослідження – моделі, методи машинного навчання та
допоміжні алгоритми прогнозування результативності тенісних ігор. Результа-
том дослідження є визначення найкращої стратегії тенісного беттингу.
Ключові слова: прогнозування результатів тенісних ігор, машинне навчання,
спортивний беттинг, стратегії ставок.
|
| id | journaliasakpiua-article-343065 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-11-09T02:11:03Z |
| publishDate | 2025 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/c1/81b91f2a479ae4337f01f55e2402bfc1.pdf |
| spelling | journaliasakpiua-article-3430652025-11-09T00:01:30Z Analysis and forecasting of the financial benefit for the tennis match outcomes by machine learning methods Прогнозування результатів тенісних матчів і аналіз фінансових вигод Shum, Kyryl Kuznietsova, Nataliia forecasting machine learning betting strategies financial benefit прогнозування результатів тенісних ігор машинне навчання спортивний беттинг стратегії ставок Tennis is one of the most popular sports in the world, attracting considerable attention from casual fans and professional analysts. The application of machine learning methods enables the accurate prediction of match results, opening up opportunities for profit through betting on likely winners. This study evaluates the financial benefits of predicting tennis match outcomes by identifying an effective sports betting strategy. The study examines various machine learning methods and auxiliary algorithms, comparing them to select the best betting strategy for maximizing the user’s potential profit. In the paper, the method and algorithm for determining effective sports betting strategies were developed. This algorithm and method were tested on tennis game datasets (for both women and men), and the best tennis betting strategy was identified. As part of the study, a software product has been developed to predict the outcomes of tennis matches. Реалізовано програмний продукт, який дозволяє прогнозувати результати тенісних матчів, розроблено метод визначення ефективних стратегій спортивних ставок. Теніс є одним із найпопулярніших видів спорту у світі, який привертає значну увагу як звичайних уболівальників, так і професійних аналітиків. Використання методів машинного навчання дає змогу ефективно прогнозувати результати матчів, що відкриває можливості для отримання прибутку від ставок на ймовірних переможців. Мета дослідження – оцінювання фінансової вигоди від прогнозування результатів тенісних ігор через пошук ефективної стратегії спортивних ставок. Розглянуто різні методи машинного навчання і допоміжні алгоритми та виконується їх порівняння з метою вибору найкращої стратегії укладання ставок задля максимізації потенційного прибутку користувача. Об’єкт дослідження – прогнозування результативності тенісних матчів. Предмет дослідження – моделі, методи машинного навчання та допоміжні алгоритми прогнозування результативності тенісних ігор. Результатом дослідження є визначення найкращої стратегії тенісного беттингу. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2025-09-29 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/343065 10.20535/SRIT.2308-8893.2025.3.07 System research and information technologies; No. 3 (2025); 87-98 Системные исследования и информационные технологии; № 3 (2025); 87-98 Системні дослідження та інформаційні технології; № 3 (2025); 87-98 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/343065/330998 |
| spellingShingle | прогнозування результатів тенісних ігор машинне навчання спортивний беттинг стратегії ставок Shum, Kyryl Kuznietsova, Nataliia Прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| title | Прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| title_alt | Analysis and forecasting of the financial benefit for the tennis match outcomes by machine learning methods |
| title_full | Прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| title_fullStr | Прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| title_full_unstemmed | Прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| title_short | Прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| title_sort | прогнозування результатів тенісних матчів і аналіз фінансових вигод |
| topic | прогнозування результатів тенісних ігор машинне навчання спортивний беттинг стратегії ставок |
| topic_facet | forecasting machine learning betting strategies financial benefit прогнозування результатів тенісних ігор машинне навчання спортивний беттинг стратегії ставок |
| url | https://journal.iasa.kpi.ua/article/view/343065 |
| work_keys_str_mv | AT shumkyryl analysisandforecastingofthefinancialbenefitforthetennismatchoutcomesbymachinelearningmethods AT kuznietsovanataliia analysisandforecastingofthefinancialbenefitforthetennismatchoutcomesbymachinelearningmethods AT shumkyryl prognozuvannârezulʹtatívtenísnihmatčívíanalízfínansovihvigod AT kuznietsovanataliia prognozuvannârezulʹtatívtenísnihmatčívíanalízfínansovihvigod |