Access to this full-text is provided by Springer Nature.
Content available from BMC Sports Science Medicine and Rehabilitation
This content is subject to copyright. Terms and conditions apply.
R E S E A R C H Open Access
Technical and tactical diagnosis model of
table tennis matches based on BP neural
network
Wenwen Huang
1
, Miaomiao Lu
2
, Yuxuan Zeng
1
, Mengyue Hu
2
and Yi Xiao
1*
Abstract
Background: The technical and tactical diagnosis of table tennis is extremely important in the preparation for
competition which is complicated by an apparent nonlinear relationship between athletes’performance and their
sports quality. The neural network model provides a high nonlinear dynamic processing ability and fitting accuracy
that may assist in the diagnosis of table tennis players’technical and tactical skill. The main purpose of this study
was to establish a technical and tactical diagnosis model of table tennis matches based on a neural network to
analyze the influence of athletes’techniques and tactics on the competition results.
Methods: A three-layer Back Propagation (BP) neural network model for table tennis match diagnosis were
established. A Double Three-Phase evaluation method produced 30 indices that were closely related to winning
table tennis matches. A data sample of 100 table tennis matches was used to establish the diagnostic model
(n= 70) and evaluate the predictive ability of the model (n=30).
Results: The technical and tactical diagnosis model of table tennis matches based on BP neural network had
a high-level of prediction accuracy (up to 99.997%) and highly efficient in fitting (R
2
= 0.99). Specifically, the
technical and tactical diagnosis results indicated that the scoring rate of the fourth stroke of Harimoto had
the greatest influence on the winning probability.
Conclusion: The technical and tactical diagnosis model of table tennis matches based on BP neural network
was highly accurate and efficiently fit. It appears that the use of the model can calculate athletes’technical
and tactical indices and their influence on the probability of winning table tennis matches. This, in turn, can
provide a valuable tool for formulating player’s targeted training plans.
Keywords: Artificial neural network, Table tennis, Techniques and tactics, Diagnostic model, Winning probability
Background
Compared with other physical events (such as track and
field, cycling, swimming, etc.), one of the most important
characteristics of ball games is that there is a nonlinear
relationship between athletes’performance and their
sports quality [1]. When sports are classified into five
levels according to their technical and tactical import-
ance, and ball games are highest on the list as a success-
ful outcome is predicated on the combined use of
techniques and tactics [2]. Therefore, being able to diag-
nose athletes’technical and tactical performance is ex-
tremely important not only for training but preparation
for competition.
Table tennis is very popular and suitable for most
people. The current table tennis match is an 11-point
system, with the two sides serving alternately and two
© The Author(s). 2021 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if
changes were made. The images or other third party material in this article are included in the article's Creative Commons
licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons
licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the
data made available in this article, unless otherwise stated in a credit line to the data.
* Correspondence: cutexxx@163.com
1
China Table Tennis College, Shanghai University of Sport, Shanghai 200438,
China
Full list of author information is available at the end of the article
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54
https://doi.org/10.1186/s13102-021-00283-3
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
serves per round. Compared with other ball games, table
tennis has the characteristics of many variations, fast ball
speed, and fast spin speed, which requires a high skill
level for professional table tennis players. The outcomes
of table tennis competitions are affected by multiple fac-
tors, such as physical fitness, psychology, techniques, and
tactics. Perhaps the most influential of these are tech-
niques and tactics which appear to have a direct effect on
the competition results [3]. The technical and tactical
diagnosis of table tennis competitions can not only pro-
vide useful guidance for training but also improve the
competitive ability of athletes [4,5]. Therefore, the sys-
tematic analysis and diagnosis of the players’technical and
tactical characteristics are very important and crucial in
preparing athletes for table tennis competitions [6,7].
With regard to the technical and tactical diagnosis
methods of table tennis, the most well-known is Wu and
colleagues’three-phase evaluation method [8]. This ap-
proach uses a player’s last stroke in each rally as the ob-
servation unit and divides each rally into three phases:
(a) attack after service (the first and third strokes), (b) at-
tack after receive (the second and fourth strokes), and
(c) stalemate phase (after the fourth strokes). The scoring
rate (SR) and the usage rate (UR) are used as evaluation
indices for analysis. This method has been widely used in
many studies to evaluate the strength of table tennis
players’techniques and tactics. Practically, the top table
tennis teams such as the People’s Republic of China’sna-
tional team uses this method to conduct technical and
tactical analyses for their own players and major oppo-
nents while preparing for the World Championships and
Olympic Games [7,9–11].
A table tennis game can be divided into three stages
according to the time sequence characteristics of a game:
(a) beginning, (b) middle, and (c) end. In 2018, Xiao and
his colleagues posited that a more detailed analysis may
result by combining the three-phase evaluation method
and the three stages of a game [12]. Calling it the Double
Three-Phase, this approach still uses the SR and UR in-
dices but calculates them for the beginning, middle, and
end stages of each game. Thus, coaches and athletes are
provided a more fine-grain analysis to evaluate the
strength of a player’s technical and tactical performance
in terms of variation of use and at what points in the
match [12]. This provides a more comprehensive ana-
lysis of the variation and strength of players’techniques
and tactics in the competition, which is a useful supple-
ment to the three-phase evaluation method.
With the development of information technology,
there are a variety of diagnosis and analysis methods for
game performance, such as the Multiple Liner Regres-
sion analysis, the Markov Chain, and the Artificial
Neural Network (ANN) [7]. Given that table tennis
matches are affected by many factors, and the
relationship between the competition result and players’
technical and tactical ability is mainly nonlinear [13], the
use of these linear analyses methods makes it difficult to
determine the impact of athletes’technical and tactical
performance on the result. Therefore, the two-fold pur-
pose of this study was to: (a) develop a computational
model based on a nonlinear model that can simulate the
winning probability of a table tennis match, and (b)
apply the model to analyze the technical and tactical
diagnosis of a world’s elite table tennis player based on
the Double Three-Phase evaluation method. Through
changing of index weights in the diagnostic model, the
players’strengths and weaknesses in techniques and tac-
tics could be found, which provided a valuable reference
for coaches to formulate targeted training plans and to
adjust the technical and tactical strategy in matches, and
for players, it is helpful to overcome the shortcomings in
techniques and tactics and improve their table tennis
technique levels.
Literature review
A Multiple Linear Regression analysis model can be used
to diagnose the relationship between sports competition
performance and a series of influencing factors [14].
One advantage of a multiple line regression model is
that it can better reveal the impact of each independent
variable on the dependent variable. However, the multiple
line regression analysis model requires close linear correl-
ation between independent variables and dependent vari-
ables, therefore, making it difficult to diagnose technical
and tactical use in table tennis matches [15].
The Markov Chain model can simulate the winning
probability of a table tennis match. When one of the in-
dices affecting the win of a match changes, the model
can diagnose the winning probability of a match again.
Hence, researchers can determine what influence a
change in one index value (increasing or decreasing) will
have on the output (winning probability) by comparing
the difference of winning probability. Pfeiffer et al. pro-
posed four different state transition models to describe
the tactical behavior of table tennis by using the Markov
Chain, and the results demonstrated that the random
path simulation of the Markov Chain was valuable in the
technical and tactical diagnosis of table tennis matches
[16]. Newton used the stochastic Markov Chain model
to propose the probability density function of players’
winning in tennis matches, and the results showed that
the Markov Chain could provide more detailed informa-
tion for the diagnosis of players’athletic ability and the
winning probability of tennis matches [7,17].
An Artificial Neural Network (ANN) is a complex net-
work system that carries out parallel processing and non-
linear transformation of information by simulating human
brain neural processing. Compared with traditional
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 2 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
multiple linear regression, ANN has a high nonlinear dy-
namic processing ability [18]. The basic idea of a neural
network model is quite similar to that of the Markov
Chain model but the latter requires the integrity of data
structure and its time sequence while the ANN has no
strict requirement for input indices [7]. When the input
and output relationship is too complicated to be expressed
by general formulas, it is easy for neural networks to
realize their highly nonlinear mapping relationship. The
learning process of a neural network is essentially a
process of optimizing the weights of the neural network,
which often requires hundreds of iterative operations.
Hence, choosing a reasonable neural network structure
can reduce network training times and improve the train-
ing accuracy of the model. ANNs have been widely used
in the fields of pattern recognition, automatic control, and
economic diagnosis, etc. [19–22].
A possible option to these three analysis methods is a
Back Propagation (BP) neural network. A BP neural net-
work is a multi-layered feed forward neural network
trained by backward error propagation [23]. This algo-
rithm proposed by Rumelhart and McClelland uses
squared error as its objective function and calculates the
minimum of the error function by gradient descent
method [7]. The basic structure of a three-layer BP
neural network includes three layers: (a) input, (b) hid-
den, and (c) output. The key to establishing a three-layer
BP network model is the determination of the input
layer’s neuron number, the hidden layer’s number, the
hidden layer’s neuron number, the selection of transfer-
ring function as well as the training function. The high
nonlinear processing ability of BP neural network makes
it widely applied in game prediction, and techniques and
tactics diagnosis for ball matches [13,24,25]. Yang &
Zhang used BP neural network and multiple regression
to analyze the technical and tactical ability of elite male
table tennis players, and the results showed that the
overall fitting accuracy of the BP neural network was
higher than that of the multiple regression model [14].
At present, some potential innovative approaches have
been proposed and applied in the outcome prediction of
other sports, e.g. applying passing network indicators to
predict the winning probability of football games, or
introducing the negative Poisson binomial distribution
to assess the probabilities of all possible outcomes of a
darts tournament [26,27].
Given that the BP neural network model has no strict
requirement for input indices and has high fitting accur-
acy, the use of the BP neural network seems appropriate
when attempting to establish a technical and tactical
diagnosis model of table tennis matches. When one of
the technical and tactical index’s value is adjusted, the
winning probability of the match can be recalculated
through the diagnostic model in order to examine the
impact of that index on the winning probability. Thus,
this model was considered to be applied in this study.
Method
Three-phase evaluation method
The Three-Phase evaluation method of table tennis
technique and tactic divides each rally into three phases:
the serve and attack phase, the receive and attack phase,
and the stalemate phase. It uses the scoring rate (SR)
and the usage rate (UR) in each phase of each game to
analyze the strength of table tennis players’techniques
[7]. The definition of SR and UR were as follows [4]:
Usage Rate: URi¼WiþLi
S
Scoring Rate: SRi¼WiþLi
S
where iis the number of each phase, UR
i
is the usage
rate at the i-th phase, SR
i
is the scoring rate at the i-th
phase, W
i
is the scoring points at the i-th phase, and L
i
is the losing points at the i-th phase, and S is the total
points of both sides scored during a match.
Division of a game
According to the time sequence characteristics of each
game, a table tennis game could be divided into three
stages: the beginning, the middle, and the end, which
were defined as follows: (a) the beginning of a game: be-
fore the score of either side reaches 5 points (first to
fourth points), (b) the middle of a game: from the score
of either side reaches 5 points to 8 points (fifth to eighth
points), and (c) the end of a game: from the score of ei-
ther side reaches 9 points to the end of a game.
Double three-phase evaluation method
Combining the Three-Phase evaluation method with the
division of a game, Xiao et al. put forward the Double
Three-Phase evaluation method, which still used the SR
and UR in the beginning, middle, and the end of a game
to evaluate the strength of players’technical and tactical
performance [12].
The analysis indices based on the double three-phase
evaluation method
In this study, the 30 technical and tactical analysis indi-
ces which have a closer relation to winning a competi-
tion were selected based on the Double Three-Phase
evaluation method, including three kinds of scoring rate
and usage rate, which were illustrated in Table 1.
Definition of the competition result
In this paper, the player’s competition result of a match
was described by the winning probability, which was de-
fined as the total winning points of one side divided by
the total scores of both sides in a match [6].
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 3 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
The calculation of the weights of technical and tactical
indices on winning probability
In this study, the impact of technical and tactical indices
on the winning probability was examined by adjusting
the value of each technical and tactical index, while
keeping the other indices’value unchanged, and recalcu-
lating the winning probability of the match through the
diagnostic model [7].
The calculation steps of the weight of each technical
and tactical index’s impact on the winning probability
were as follows:
(1) Inputting the unchanged data matrix into the
neural network diagnostic model to training, and
calculating the winning probability (wp1) of the
matches.
(2) According to the Formula (1), adjusting the value of
the selected technical and tactical index, and keep
the others unchanged [28].
y¼0:238cos ‐1:32xþ0:66ðÞ‐0:178 ð1Þ
where x is the unchanged value of the selected index,
and y is the adjusted increment.
(3) Inputting the changed data matrix into the neural
network diagnostic model to training again, and
recalculating the winning probability (wp2) of the
matches.
(4) Calculating the weight of selected technical and
tactical index on the winning probability:
weight = (wp1 ‐wp2)/wp1 ∗100%. The larger the
absolute weight value is, the higher the
influence of the selected index on the winning
probability.
Neural network diagnostic model
Neural network model structure In this study, 100
matches of the world’s elite male table tennis players
were selected as sample data. The selection criteria were:
(a) ITTF world matches held from the year 2018 to
2020, (b) men’s singles, (c) elite table tennis player, and
(d) the game video is complete and can play normally.
The technical and tactical analysis of these matches was
conducted by the research team using the Double
Three-Phase evaluation method.
The descriptive statistics results of the 30 technical
and tactical indices of these matches were shown in
Table 2. The results of the technical and tactical analysis,
including the 30 technical and tactical analysis indices
and the winning probability, were used as the data sam-
ple, and a three-layer BP neural network model for table
tennis match diagnosis was established by adopting the
Levenberg-arquardt training function.
The input layer of the diagnostic model included 30
neurons which corresponded to 30 technical and tactical
analysis indices (from × 1 to × 30), respectively. The out-
put layer of the diagnostic model was the winning prob-
ability of a table tennis match, and the hidden layer
included 31 neurons. The number of the hidden layer’s
neurons can be changed according to the requirement of
the model’s accuracy. The nonlinear S transferring func-
tion (activation function) tansig(n) was taken between
the input layer’s neurons and the hidden layer’s neurons,
and the linear transferring function purelin(n), i.e. f(x) =
x, between the hidden layer’s neurons and the output
layer’s neurons was taken, whose structure was shown in
Fig. 1[29].
Neural network model training The 100 matches were
randomly divided into two groups, of which 70 matches
were taken as neural network training set to establish
Table 1 The 30 technical and tactical analysis indices based on the double three-phase evaluation method
The beginning of a game The middle of a game The end of a game
Phases Bh Name Bh Name Bh Name
serve and attack phase X1 UR of Serve X3 UR of Serve X5 UR of Serve
X2 SR of Serve X4 SR of Serve X6 SR of Serve
X13 UR of the third stroke X15 UR of the third stroke X17 UR of the third stroke
X14 SR of the third stroke X16 SR of the third stroke X18 SR of the third stroke
receive and attack phase ×7 UR of Receive X9 UR of Receive X11 UR of Receive
X8 SR of Receive X10 SR of Receive X12 SR of Receive
X19 UR of the fourth stroke X21 UR of the fourth stroke X23 UR of the fourth stroke
X20 SR of the fourth stroke X22 SR of the fourth stroke X24 SR of the fourth stroke
stalemate phase X25 UR after the fourth strokes X27 UR after the fourth strokes X29 UR after the fourth strokes
X26 SR after the fourth strokes X28 SR after the fourth strokes X30 SR after the fourth strokes
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 4 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
the diagnostic model of table tennis matches, and the
other 30 matches were taken as the test set to evaluate
the predictive ability of the diagnostic model.
The BP self-learning algorithm was used to learn the
training samples. After learning, the winning probability
(the output data) of the table tennis diagnostic model
can be obtained. Then, the input and output data of the
diagnostic model can reflect the functional relationship
between the technical and tactical indicators (input
layer) and the winning probability (output layer) of the
matches. In order to make the diagnostic model have
faster convergence speed and higher stability, the
Levenberg-Marquardt training function was adopted in
this paper. The Levenberg-Marquardt training rule had
the characteristics of fast calculation speed and high pre-
cision, whose specific steps were as follow [29]:
Step-1: Setting initial weight W;
Step-2: For the input indices (×
1
,×
2
...... ×
30
) of all
training samples: (a) Calculating the output of hidden
neurons: Hj = f (∑Wji Xi ‐θj) (i = 1 to 30; j = 1 to 31),
where W
ji
is the connection weight between neuron i
and j, θ
j
is the j-th hidden neuron’s threshold value, (b)
Calculating the output of the output neuron: y =
f(∑Wj Hj ‐θo) (j = 1 to 31), where W
j
is the
connection weight between neuron j and output
neuron, θ
o
is the output neuron’s threshold value, and
(c) Calculating the mean squared error of all sample’s
actual and expected output: e = ∑(t ‐y)
2
/2 , where y is
the actual output value, and t is the expected output
value.
Step-3: Anti-propagation computing weight correction
vector: Δw=[J
T
J + uI]
‐1
J
T
e; Correcting all net weights:
wk+1=wk+Δw, where k is training steps, J is
Jacobian matrix derived by the partial derivative of the
error to weight, I is a unit matrix, e is network error
vector, and u is the increasing coefficient (a scalar).
When the error has not reached the expected value, u
is reduced, otherwise, u is increased.
Step-4: Repeating Step-2 and Step-3 until one of the
following conditions is met: the error meets the re-
quirements or the training step exceeds the given value.
Neural network model programming The training
process of a neural network is essentially a process of
optimizing the weights of the neural network, which re-
quires hundreds of iterative operations, so the amount
of calculation in the training process of the neural
Table 2 Descriptive statistics of the 30 technical and tactical indices of the 100 matches
Variables M ± SD Variables M ± SD Variables M ± SD
x1 .0645 ± .05043 x11 .1557 ± .0683 x21 .1973 ± .0607
x2 .7663 ± .4021 x12 .6379 ± .2378 x22 .4234 ± .2055
x3 .0620 ± .0458 x13 .2005 ± .0662 x23 .1975 ± .0611
x4 .8001 ± .3613 x14 .5617 ± .2431 x24 .3792 ± .1969
x5 .0567 ± .0400 x15 .2038 ± .0620 x25 .3917 ± .1060
x6 .8093 ± .3575 x16 .5376 ± .1879 x26 .4038 ± .1669
x7 .1577 ± .0661 x17 .2234 ± .0662 x27 .3861 ± .0957
x8 .6065 ± .2771 x18 .5527 ± .1813 x28 .3988 ± .1497
x9 .1507 ± .0583 x19 .1857 ± .0576 x29 .3667 ± .1079
x10 .6494 ± .2339 x20 .3912 ± .2355 x30 .3917 ± 1426
Fig. 1 Neural network model structure
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 5 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
network is quite large and needs to be realized by pro-
gramming. The neural network development kit of
Matlab R2016b was utilized to develop the diagnostic
model’s program. The pseudocode frame for the net-
work training procedure was shown as following:
Input:
Input data to present to the network /* inputs of the
model */
Target data defining network output /* targets of the
model */
Randomly divide up the samples
Setting:
Input the number of the hidden neurons
Choosing a training algorithm
Training:
Training the model until meeting the requirement of
the error setting
Output
Results
The neural network diagnostic model was evaluated via
the leave-one-out cross validation method, which is
commonly used when dealing with a small number of
samples [30].
Training performance
The training time (epochs) and mean squared error
(MSE) of the table tennis diagnostic model were shown
in Fig. 2. The x-axis is the training times (epochs) and
the y-axis is the training error (MSE). The goal of the
training error of the diagnostic model was set 1e-5. As
shown in Fig. 2, when the training time reached 6
epochs, the model’s training error (3.57e-7) reached and
exceeded the requirement (1e-5), which indicated the
diagnostic model’s training performance was high.
Goodness-of-fit of the model After training and learn-
ing, the SIM function (program code: t_sim = sim (net,
a)) was used to simulate the output of the model, and
compare the output data with the observed values (tar-
gets) to evaluate the predictive ability of the diagnostic
model. The result was shown in Fig. 3, in which the R
value represented the correlation between outputs and
targets (1 means a close relationship and 0 a random re-
lationship), and the output data points (small white cir-
cle) of the neural network diagnostic model were all
close to the dotted line, indicating that the network had
a good performance and high goodness-of-fit (R
2
= 0.99).
The prediction accuracy of the model The 30 index
data of the other 30 matches were input as test samples
into the neural network diagnostic model of table tennis
matches, and the simulated values of the winning prob-
ability of these matches were obtained. The maximum
absolute error, the minimum absolute error, and the
mean absolute error were calculated by comparing the
simulated values of the model with the observed (real)
values. These values were used to examine the model’s
prediction accuracy, and the results were shown in
Table 3and Fig. 4.
The data of Table 3showed that the maximum abso-
lute error of the 30 matches was about 0.0016, while the
minimum error was about 0.000006, the average error of
the 30 matches was 0.0004. As shown in Fig. 4, the
Fig. 2 Training time and error
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 6 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
winning probability derived from the BP neural network
diagnostic model of table tennis match was very close to
the actual value, which showed that the diagnostic
model based on BP neural network reached a high pre-
diction accuracy up to 99.997% [14].
Application of the diagnostic model
In this paper, the Japanese famous table tennis player,
Harimoto, was taken as an example. 20 international
matches that Harimoto participated in in the past 5 years
(from 2015 to 2019) and the technical and tactical ana-
lysis of the 20 matches was conducted using the double
three-phase evaluation method. Then, the 30 technical
and tactical analysis indices were input into the neural
network diagnostic model for training and the outcome
(winning probability) of the matches was obtained. The
weights of 30 technical and tactical indices’impact on
the winning probability were also calculated by adjusting
the value of each technical and tactical index one by one
according to the Formula (1) and keep the others un-
changed (repeat 30 times in total).
The results of the technical and tactical diagnosis analysis
based on BP neural network model were shown in Fig. 5.
The perspective of game division (the beginning, the
middle, and the end) At the beginning of the game, the
weight of × 20 was the largest, indicating that the SR of
the fourth stroke had the greatest influence on the
winning probability, followed by × 2, × 8, and × 14,
which showed that the SR of the serve, receive, and the
third stroke had a greater influence on the winning
probability. The two technical and tactical indicators ×
13 and × 19 had a certain impact on the winning prob-
ability. The indicators × 1, × 7, × 25, and × 26 had little
effect on the winning probability.
In the middle of the game, the weight of × 22 was the
largest, indicating that the SR of the fourth stroke had
the greatest impact on the winning probability, followed
by × 4, × 10, and × 16, which showed that these three in-
dicators had a greater influence on the winning probabil-
ity. The two technical and tactical indicators × 15 and ×
21 had a certain impact on the winning probability. The
indicators × 3, × 9, × 27, and × 28 had little effect on the
probability of winning.
At the end of the game, the weight of × 24 was the lar-
gest, indicating that the SR of the fourth stroke had the
largest influence on the winning probability, followed by
× 6 and × 18, which showed that the SR of serve and the
third stroke had a greater influence on the winning
probability. The three technical and tactical indices × 12,
× 17, and × 23 had a certain influence on the winning
probability. The indicators × 5, × 11, × 29, and × 30 had
little impact on the winning probability.
The perspective of the division of the three-phase
evaluation method At the serve and attack phase, the
Fig. 3 Goodness-of-fit of the model
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 7 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
weight values of X2 and × 14 were large, indicating that
the SR of the serve and the third stroke at the beginning
of the game had the greatest influence on the winning
probability, followed by × 4, × 6, × 16, and × 18, which
showed that these four technical and tactical indicators
had a greater impact on the probability of winning. The
indicators × 13, × 15, and × 17 had a certain influence on
the winning probability. The indicators of X1, × 3, and
× 5 had little effect on the winning probability.
At the receive and attack phase, the weight of × 20 was
the largest, indicating that the SR of the fourth stroke at
the beginning of the game had the greatest influence on
the winning probability, followed by × 22 and × 24,
which showed that the SR of the fourth stroke in the
middle and at the end of the game had a greater influ-
ence on the winning probability. The indicators × 8, ×
10, × 12, × 19, × 21, and × 23 had a certain influence on
the winning probability. The indicators X1, × 9, and × 11
had little effect on the winning probability.
At the stalemate phase, all the indicators × 25, × 26, ×
27, × 28, × 29, and X30 had little effect on the winning
probability.
Discussion
The Double Three-Phase evaluation method of table
tennis matches provides a more scientific and compre-
hensive perspective to analyze the competitive ability of
athletes compared with the traditional three-phase
evaluation method. The results of technical and tactical
statistics, based on the Double Three-Phase method, are
clearer and closer to the actual situation of table tennis
competition, which better reflects the variation and
strength of players’techniques and tactics in the compe-
tition [12]. The results from this study indicated that
using the Double Three-Phase evaluation method could
produce a technical and tactical diagnosis model of table
tennis matches employing a BP neural network and the
model showed high prediction accuracy. Predicting with
the generalized linear models is much simple, conveni-
ent, and easy to implement and understand, however,
the fitting degree and prediction accuracy of the estab-
lished model are not as good as that of the model based
on BP neural network in table tennis, as that there is a
nonlinear relationship between table tennis players’
sports quality and the competition results [7].
After establishing the diagnosis model, according to
the requirements of the athlete’s technical diagnosis, a
certain index value can be increased or reduced, then,
input it into the neural network diagnosis model, the
corresponding output value (winning probability) of the
model can be obtained. Thus, it is easy to find the fac-
tors that have a significant impact on the competition
result of a match and provides a theoretical reference for
formulating a targeted training plan [29]. In this study,
the diagnostic model was used to diagnose the tech-
niques and tactics of Japanese famous table tennis player
(Harimoto), and the results suggested that he should
make full use of his advantage of active attack in the
competition, and improve the quality of stalemate balls
to make himself in a relatively dominant position and
improve the winning probability, which supports the
study of Chien et al. [31].
Although the technical and tactical diagnosis model of
table tennis matches based on BP neural network had a
high prediction accuracy and had a good ability to
analyze the technical and tactical abilities of table tennis
players, there still may be some difference between the
output of the model and the true competition results.
Several probable factors that may contribute to the dif-
ference were the: (a) training and test sample size. In this
Table 3 The error between actual and simulated values
Match No Simulated value Actual value Error
No.1 0.452194 0.452200 0.000006
No.2 0.514612 0.514600 0.000012
No.3 0.333274 0.333300 0.000026
No.4 0.549943 0.550000 0.000057
No.5 0.474432 0.474500 0.000068
No.6 0.449924 0.450000 0.000076
No.7 0.426923 0.427000 0.000077
No.8 0.437415 0.437500 0.000085
No.9 0.466303 0.466200 0.000103
No.10 0.500112 0.500000 0.000112
No.11 0.402319 0.402200 0.000119
No.12 0.619540 0.619700 0.000160
No.13 0.525672 0.525500 0.000172
No.14 0.598103 0.597900 0.000203
No.15 0.464377 0.464600 0.000223
No.16 0.419550 0.419800 0.000250
No.17 0.463129 0.462800 0.000329
No.18 0.535736 0.535400 0.000336
No.19 0.464351 0.464000 0.000351
No.20 0.458779 0.459200 0.000421
No.21 0.554037 0.554600 0.000563
No.22 0.534597 0.534000 0.000597
No.23 0.300882 0.301600 0.000718
No.24 0.460113 0.459300 0.000813
No.25 0.454538 0.455400 0.000862
No.26 0.535627 0.536600 0.000973
No.27 0.536156 0.537200 0.001044
No.28 0.464899 0.466000 0.001101
No.29 0.438844 0.440000 0.001156
No.30 0.524326 0.525900 0.001574
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 8 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
study, the training and test sample size was 70 and 30
respectively, which can be still increased for a higher
model’s prediction accuracy [18], (b) selection of tech-
nical and tactical diagnosis indices. In the study, only 30
technical and tactical analysis indices were selected and
the data were input into the model to calculate the
output. There are also other factors influencing the
player’s competition results such as psychological and
environmental elements, the opponent, and so on [32],
(c) the structure of the neural network model, including
the number of hidden layers, the number of the hidden
layer’s neurons, the training and transferring function,
Fig. 4 Accuracy of the diagnostic model
Fig. 5 The technical and tactical diagnostic results of Harimoto
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 9 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
et al., which can all be changed according to the require-
ment of the model’s accuracy [33], and (d) the indicator
winning probability can only partly reflect the competi-
tion result of a match. Hence, the output of the diagnos-
tic model cannot fully represent the actual competition
result.
It appears that the BP neural network can retain the
mapping relationship between the index and the past
competition result, and the technical and tactical level of
athletes will change over time. When the technical and
tactical level of the athlete changes, as long as the latest
technical and tactical data are input into the BP neural
network model, the new connection weights that reflect
the current technical and tactical level of athletes can be
obtained by retraining the existing neural network
model. Therefore, it is necessary to track the athletes’
techniques and tactics for a long time to collect new
technical and tactical data to retrain the model, thus the
diagnostic model can be modified and maintain a good
diagnostic ability.
Conclusions
The technical and tactical diagnosis model of table tennis
matches based on BP neural network had a high predic-
tion accuracy and highly efficient in fitting. By using this
model, the weights of the influence of athletes’technical
and tactical indices on the winning probability of the com-
petition can be calculated, which provides a valuable refer-
ence for formulating targeted training plans of players.
For Harimoto, the technique of attack after receive has
the greatest influence on the probability of winning,
followed by the attack after serve. The stalemate tech-
nique has little effect on the winning probability.
Implications
The diagnosis model can be used to analyze the tech-
nical and tactical characteristics of table tennis players
and evaluate the strength of athletes’techniques and tac-
tics. After the weights of the influence of athletes’tech-
nical and tactical indices on the winning probability of
the competition are calculated, the indices that have a
significant impact on the winning probability can be de-
termined, which helps coaches to make targeted training
and competition plan for table tennis players to improve
the athletic ability of players.
For Harimoto, he should strengthen the training of the
attack after serve and attack after receive skills, strengthen
the consciousness of attack, and improve the quality of
stalemate balls so as to increase the scoring rate and en-
hance the winning probability of competition.
Limitations
In this study, only 100 matches of the world’s elite male
table tennis players were selected. In the follow-up
study, more matches can be selected as training samples
to gain a better training and fitting accuracy of the BP
neural network model. Besides, this study took only one
player as an analyzing example, more players from dif-
ferent countries can be selected for a comparative ana-
lysis of their technical and tactical characteristics in the
future study. Moreover, the technical and tactical indices
did not involve specific technical action, so the sugges-
tion proposed may not be targeted enough. Finally, the
tests to confirm the linearity/non-linearity relationship
between the competition results and players’sports qual-
ity can be conducted in the future study.
Supplementary Information
The online version contains supplementary material available at https://doi.
org/10.1186/s13102-021-00283-3.
Additional file 1: Appendix I. The Matlab code for the evaluation of
the diagnostic model. Appendix II. The Matlab code for the application
of the model (“Harimoto”case).
Acknowledgments
Not applicable.
Authors’contributions
All authors read and approved the final manuscript. Conceptualization, Y.X.
and W.H.; Methodology, Y.X. and W.H.; Tool, Y.X.; Data curation, W.H., Y.X., and
Y.Z.; Formal analysis, Y.X., W.H., and Y.Z.; Writing-original draft preparation, Y.X.,
M.L., Y.Z., and M.H.; Writing-review and editing, W.H., M.L., Y.Z., and M.H.; Supervi-
sion, Y.X.
Funding
This study was funded by the Shanghai Science and Technology
Commission (No.18080503100).
Availability of data and materials
The datasets generated during the current study are not publicly available,
but are available from the corresponding author upon reasonable request.
Declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Author details
1
China Table Tennis College, Shanghai University of Sport, Shanghai 200438,
China.
2
School of Economics and Management, Shanghai University of Sport,
Shanghai, China.
Received: 26 January 2021 Accepted: 10 May 2021
References
1. Hohmann A, Lames M, Letzelter M. Einführung in die Trainingswissenschaft.
Wiebelsheim: Limpert; 1999.
2. Tian MJ, Wu FQ. The exploration of making sports training scientific. Beijing:
People’s Sports Publishing Press; 1988.
3. Kong QY, Li XQ. Prediction method of table tennis match based on
condition number and genetic-neural network. J Guilin Univ Aerosp
Technol. 2015;20(4):528–32.
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 10 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
4. Fuchs M, Liu R, Malagoli Lanzoni I, Munivrana G, Straub G, Tamaki S. Table
tennis match analysis: a review. J Sports Sci. 2018;36(23):2653-62.
5. Guan YP, Ye Y, Li JJ, Si J, Zhang H, editors. Skill and tactic analysis for table
tennis matches. Nanjing: International Conference on Computer Science
and Service System (CSSS); 2011.
6. Mao W, Yu L, Zhang H, Wang H, Wang J, editors. Skill and tactic diagnosis
for table tennis matches based on artificial neural network and genetic
algorithm. Sixth International Conference on Natural Computation. 2010.
7. Zhang H, Zhou Z, Yang Q. Match analyses of table tennis in China: a
systematic review. J Sports Sci. 2018;36(23):2663-74.
8. Wu H, Wu HO, Cai X, Sun Q, Wang S, Chen M. Strength evaluation and
technique analysis on Chinese players in table tennis matches of the 1988
Olympic games. China Sport Sci Technol. 1988;(8):21–9.
9. Huang W, Shi Z. Three-stage index evaluation about Ding Ning’s table
tennis playing. China Sport Sci Technol. 2016;52(5):126–30.
10. Hsu MH, Chen YF, Wang SC. Offense - defense mode analysis of the world
top male table tennis player - a case study by Chuang Chi-Yuan who
participated in 2012 London Olympic male single games. J Sci Innov. 2014;
4(2):41–50.
11. Hsu MH. A study on the technical analysis and attack-defense performance
of men’s top four single players in 2008 Olympic games. Int J Table Tennis
Sci. 2010;6:248–60.
12. Xiao DD, Zhou XD, Liu H, Qin ZJ, Yu Y. The construction and application of
double three-phase method on table tennis technique and tactics. China
Sport Sci Technol. 2018;5:112–6.
13. Xie H. A study on the prediction of sports performance by using
multivariable GM (1~N) grey model. Contemp Sports Sci Technol. 2018;
8(13):221–224, 226. https://doi.org/10.16655/j.cnki.2095-2813.2018.13.221.
14. Yang Q, Zhang H. Application of BP neural network and multiple regression
in table tennis technical and tactical ability analysis. J Chengdu Sport Univ.
2016;42(1):78–82.
15. Gong J. Olympic games China men’s basketball team performance based
on artificial neural network to predict the experimental studies. Hubei:
Wuhan Sports University; 2007.
16. Pfeiffer M, Zhang H, Hohmann A. A Markov chain model of elite table
tennis competition. Int J Sports Sci Coach. 2010;5(2):205–22. https://doi.
org/10.1260/1747-9541.5.2.205.
17. Newton PK, Aslam K. Monte Carlo tennis: a stochastic Markov chain model.
J Quant Anal Sports. 2009;5(3):44.
18. Siregar SP, Wanto A. Analysis of artificial neural network accuracy using
backpropagation algorithm in predicting process (forecasting). Int J Inf Syst
Technol. 2017;1(1):34–42.
19. Basu JK, Bhattacharyya D, Kim T. Use of artificial neural network in pattern
recognition. Int J Softw Eng Appl. 2010;4(2):23-33.
20. Ghoneim SSM, Taha IB. Artificial neural networks for power transformers
fault diagnosis based on iec code using dissolved gas analysis. Int J Control
Autom Syst. 2015;4(2):18–21.
21. Syafaruddin S, Karatepe E, Hiyama T, editors. Controlling of artificial neural
network for fault diagnosis of photovoltaic array. International Conference
on Intelligent System Application to Power Systems. 2011.
22. Yang S, Li W, Wang C. The intelligent fault diagnosis of wind turbine
gearbox based on artificial neural network. International Conference on
Condition Monitoring & Diagnosis. 2008.
23. Rumelhart DE, McClelland JL. Parallel distributed processing: explorations in
the microstructure of cognition. USA: MIT Press; 1986. https://doi.org/10.
7551/mitpress/5236.001.0001.
24. Shao SH. Nonlinear correction analysis on the popular aerobics performance
based on BP neural network model. J Gansu Lianhe Univ (Nat Sci). 2012;
26(1):64–8.
25. Zhang J, Cao P, Shi F, Niu Y, Wu Y, Wang X, et al. Analysis and modeling of
football team’s collaboration mode and performance evaluation using
network science and BP neural network. Math Probl Eng. 2020;2020:1–13.
26. Ievoli R, Palazzo L, Ragozini G. On the use of passing network indicators to
predict football outcomes. Knowl-Based Syst. 2021;222:106997.
27. Liebscher S, Kirschstein T. Predicting the outcome of professional darts
tournaments. Int J Perform Anal Sport. 2017;17(5):666–83. https://doi.org/1
0.1080/24748668.2017.1372162.
28. Xiao Y, Zhang H. Research report on the preparations of Chinese table
tennis team for the Olympics - on the diagnostic model of table tennis
competition based on artificial neural network. Sport Sci Res. 2008;29(06):
19–22.
29. Zhang H. Computer diagnosis and analysis of ball games. Heilongjaing:
Heilongjiang Science and Technology Press; 2006.
30. Teixeira AFDS, Sérgio Henrique Godinho Silva TSD, Silva AO, Moreira FMDS.
Soil physicochemical properties and terrain information predict soil
enzymes activity in phytophysiognomies of the quadrilátero ferrífero region
in Brazil. Catena. 2020;199:105083.
31. Chien YC, Hsu CT, Wu SK, Hsu MH. 3s theory for case analysis: tactical
characteristics of two talented young table tennis players. J Xingda Sports.
2020;19:41–58.
32. Rossi PG. The genetic basis of human athletic performance. Why are
psychological components so often overlooked? J Physiol. 2008;586:3017.
33. Baczynski D, Parol M. Influence of artificial neural network structure on
quality of short-term electric energy consumption forecast. IEEE Proc
Generation Transm Distrib. 2004;151(2):241–5. https://doi.org/10.1049/ip-
gtd:20040070.
Publisher’sNote
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Huang et al. BMC Sports Science, Medicine and Rehabilitation (2021) 13:54 Page 11 of 11
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Available via license: CC BY 4.0
Content may be subject to copyright.
