Monitoring the Mental Status of Football Players
Elena Smets
1
, Pieter Joosen
1,2
, Joachim Taelman
2
, Vasileios Exadaktylos
1
and Daniel Berckmans
1
1
Division Measure, Model & Manage Bioresponses, KU Leuven, Kasteelpark Arenberg 30, Leuven, Belgium
2
BioRICS NV, Technologielaan 3, Heverlee, Belgium
Keywords: Heart Rate, Football, Modelling, Real-time.
Abstract: This work has tested heart rate to measure anxiety during a penalty shootout. Until now, anxiety is measured
through questionnaires, where online monitoring is not possible. Therefore there is a need for physiological
parameters to represent anxiety online. Since it is proven that the level of anxiety is a good predictor of
penalty outcome, it was hypothesised that this outcome can be estimated with heart rate and activity. To test
this hypothesis an experiment has been conducted with 54 participants (age= 23±4,54 years). They each
performed three sessions of a penalty shootout, where heart rate and activity were measured. An adapted
version of the State-Trait Anxiety Inventory was used as reference for anxiety level. The data have been
analysed using a static and dynamic approach. These resulted in parameters that were used to predict the
anxiety level and penalty performance of the participant with a multinomial logistic regression model. The
results show that 47,11% of the participants were correctly classified into three classes of anxiety. Based on
a classification into penalty performance 55,11 % of the participants were correctly classified. It can be
concluded that heart rate in combination with activity shows promising results as predictor for anxiety.
1 INTRODUCTION
In the past, a lot of research is done with the focus
on the physical aspect of sports. Today, the focus is
shifting towards the mental aspect. The reason for
this shift is that it got to researchers’ attention that
mental health can play an important role in
performance.
The mental aspects in sports are defined here as
mental influences that have an impact on the
performance of the athlete. Generally, these
influences can be divided in two main groups. These
are on the one hand perception of effort and on the
other hand feelings of anxiety.
Perception of effort means that an exercise at the
same intensity feels harder after some time (Knicker
et al., 2011). De Morree et al., (2012) state that it is
the conscious awareness of the central motor
command sent to the active muscles. This
interpretation of perception of effort is called the
corollary discharge model (Marcora, 2009). When
humans undertake action, it is preceded by brain
activity. Specifically for voluntary actions, which
are present in sports, this brain activity takes place in
the motor areas. It is this action in the central motor
system that is sensed and is reflected by perception
of effort. The second mental influence is anxiety.
One of the most important causes for presence of
anxiety is stress. Stress is defined here as an
athlete’s ability, or lack of ability, to deal with
competitive pressure (Mateo et al., 2012).
Different methods have been developed to
measure perception of effort and the level of anxiety.
These methods are mainly based on surveys. For the
measurement of perception of effort the Borg rating
scale is the oldest and the most widely used
instrument (Chen et al., 2002). This scale is a
general measure of exercise intensity (Zamunér et
al., 2011). It is an equidistant interval chain that
starts at 6 (no exertion at all) and ends with 20
(maximal exertion) (Borg, 1998). To measure the
level of anxiety the state-trait anxiety inventory
(STAI) is most commonly used. As the name
suggests, the STAI measures both state and trait
anxiety (Horikawa and Yagi, 2012). The test
consists of two forms with each 20 items. The items
are rated on a four-point Likert scale (Horikawa and
Yagi, 2012). Based on the sum of the quotations on
each item, a measure of both state and trait anxiety is
provided.
Since these inventories are quite devious and
206
Smets E., Joosen P., Taelman J., Exadaktylos V. and Berckmans D..
Monitoring the Mental Status of Football Players.
DOI: 10.5220/0004679302060213
In Proceedings of the International Congress on Sports Science Research and Technology Support (PerSoccer-2013), pages 206-213
ISBN: 978-989-8565-79-2
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
always require some time to fill out, it is the goal of
many researchers to find physical parameters that
perform as good as the questionnaires, but are easier
to measure. For perception of effort, extensive work
has already been done to find a link of the gold
standard with physical parameters (Chen et al.,
2002). Chen et al., (2002) concluded from their work
that the physiological variable that correlates best
with the Borg scale is breath rate. For anxiety
however still some work is necessary. There are
some physical and biochemical parameters identified
such as heart rate variability (Mateo et al., 2012),
blood pressure (Frazier et al., 2002), epinephrine,
norepinephrine and cortisol (Hoehn et al., 1997).
The problem with these parameters however is that
none of them can be measured online.
It is in this context that this research tries to
make a contribution. A very specific scenario in
football where there is a need for an online
measuring method of anxiety is used. Since it is
proven that the outcome of a penalty is highly
dependent on the level of anxiety of the player
(Jordet, 2009), it could be interesting for coaches to
know the level of anxiety of each player before
deciding who will take a decisive penalty. The use
of inventories would in this case be too time
consuming and clumsy. In this research, heart rate in
combination with activity is examined as a possible
physiological parameter. The activity is represented
by the acceleration signal. Since it is proven that the
level of anxiety is a good predictor of penalty
outcome (Jordet, 2009), it is hypothesised that this
outcome can be estimated based on the measurement
of heart rate and activity. If this hypothesis is
confirmed, coaches will have an interesting tool in
deciding who will take the penalties in competition.
2 MATERIAL AND METHODS
To examine the hypothesis that the outcome of a
penalty can be estimated based on the measurement
of heart rate and activity, an experiment was
conducted. In the first section an overview of the
set-up of this experiment and the materials used is
given. The experiment was approved by the Ethical
commission of the KU Leuven (6/12/2012).
Furthermore an overview of the different methods
used for the analysis is presented in the second
section.
2.1 Experimental Set-up
To test the hypothesis, an experiment was conducted
where the participants had to perform three sessions
of penalties. In the last session the anxiety was
induced. In the analysis the heart rate was compared
before and after this induction. Based on the
differences in heart rate a prediction was done
concerning the penalty outcome and anxiety level in
the different sessions.
2.1.1 Participants
Participants were chosen from the 3
rd
and 4
th
Belgian
football division. In total, three clubs participated in
the experiment. This resulted in a sample size of 54
male participants (age = 23±4,54 year). Before the
experiment started personal data of all participants
were collected, as well as their informed consent to
participate on the experiment.
2.1.2 Sensors and Questionnaires
In total three questionnaires and two sensors were
used. The first questionnaire is for personal data
collection such as age, weight, etc. The second is the
Borg rating scale for the measurement of perception
of effort. The last is an adapted version of the STAI
which was used as the gold standard to measure the
level of anxiety. An adapted version was used since
the full version consists of 20 questions. The
questionnaire was used in between three stages of
the experiment. To pose 20 questions each time
would take too long and would possibly cause the
induced anxiety to decrease. Therefore only five
questions were retained. These were ‘I feel tensed’,
‘I feel afraid’, ‘I feel certain’, ‘I feel calm’ and ‘I
feel nervous’. These five specific questions were
chosen based on the professional input of a sports
psychologist.
For the measurement of heart rate and activity
also two sensors were used. The heart rate was
measured with a Zephyr HxM sensor (Zephyr™,
Annapolis, Maryland, US) sampled at 1 Hz. For the
measurement of activity the acceleration signal was
used. This was measured with a Sony Xperia™
smartphone (Sony, Tokyo, Japan) sampled at 50 Hz.
The acceleration was measured separately in three
dimensions.
2.1.3 Set-up
All participants had to perform three penalty
sessions. First 15 training penalties, then 5 control
penalties and finally 5 induced anxiety penalties.
Before the experiment started the participant put
on the Zephyr heart rate belt and the Sony
smartphone for activity measurement.
MonitoringtheMentalStatusofFootballPlayers
207
First the participants had the chance to practice
their penalty shooting. This was done with 15
penalties in the training session. The purpose was to
shoot the penalty in one of the two holes in the net
that was hung up at the goal. This set-up is
preferable over the normal set-up with a goalkeeper,
since now the effect of the goalkeeper on success is
ruled out. These penalties could be taken freely,
without any time constriction. The participant was
told that the outcome was of no importance for the
experiment, but that later on the outcome would
become more important and he should utilise this
practicing opportunity as well as possible. After this
information was given, the adapted STAI and the
Borg rating scale were filled in and the participant
took the 15 training penalties.
After the training session a break of two minutes
took place. During this break the participant got the
information for the control condition. Here the
participant was told that he would have to take 5
penalties with a time interval of 15 seconds. He was
informed that this exercise session was important to
test and calibrate the material of the heart rate belt. It
was important that he would try to perform as well
as possible, but the outcome of the shootout would
remain confidential. The adapted STAI was filled
out after this information was given, but before the
exercise started.
After this control session a two minutes break
took place. During this break the participant got the
information for the induced anxiety condition. He
was told that the results of this last test were the only
results that would finally be of any importance.
These would be passed on to the coach, who would
use them to make a ranking for the penalty abilities
of all the players of the team. The ranking would
also be communicated later on with the other team
players. After this information was given again the
adapted STAI was filled in and, when the two
minutes break had passed, the exercise started.
Although the participants were told that their results
would be made public, this did not really happen.
The results remained confidential at any time, unless
the participant gave his consent. After all
participants had done the experiment they were
debriefed about the real purpose and they were told
that the results remain confidential.
2.2 Analysis
The goal was to predict, based on the information of
acceleration and heart rate signal, whether or not a
participant felt more anxious in the induced anxiety
session compared with the control session and if he
would score more or less penalties. The reference
level of anxiety and penalty performance for each
participant was obtained based on the responses on
the STAI and the amount of penalties scored.
Participants could either be less anxious, no
difference or more anxious in the anxiety condition
compared with the control condition. For the penalty
performance they could either score less penalties,
no difference or more penalties in the anxiety
condition compared with the control condition.
To achieve the goal different parameters were
calculated from the heart rate signal. These
parameters could then be used for prediction. To
obtain these parameters two types of analyses were
done, being a static and a dynamic analysis. Both are
followed by a statistical analysis. These were all
performed with the MATLAB R2011b (Mathworks,
Natick, MA, USA) software.
2.2.1 Static Analysis
First for every participant the data were subdivided
into three groups being the training, control and
induced anxiety group. This subdivision was based
on the beginning and end times of every session that
were written down during the experiment. Then for
each group five static parameters were calculated.
These were the mean, the maximum, the minimum,
the recovery slope and the increase slope of the heart
rate signal during each session. The recovery slope
was calculated based on the last 40 s, the increase
slope on the first 20 s of every session. The
calculation of these slopes was done using the
beginning and end point of the time frames. This
resulted in 15 values for each participant, coming
from five static parameters with each three values
for the three different sessions. These parameters did
not provide any dynamics of the heart rate, but
solely static information, hence the name of the
static analysis. The parameters were subsequently
compared among the participants. For this
comparison not the absolute values of the parameters
were used, but the pattern they followed. For each
parameter the three different values, respectively
from training, control and induced anxiety session,
were placed next to each other. These values could
increase, stay equal, form a maximum, form a
minimum or decrease. These patterns all
corresponded with a number from zero to four
respectively. This resulted in five values for each
participant, being the patterns for the five static
parameters.
icSPORTS2013-InternationalCongressonSportsScienceResearchandTechnologySupport
208
2.2.2 Dynamic Analysis
In this analysis the heart rate response of the
participants was modelled, taking the dynamics of
the heart rate into account. Both error and model
parameters were then used as parameters to predict
to which class a participant belongs. This analysis
was done with the Captain toolbox (Taylor et al.,
2007) in MATLAB.
For the modelling the Box-Jenkins methodology
was used (Zhang et al., 2012). This system is based
on the following equation
y
t
B
z
A
z
u
t
D
z
C
z
e
t
(1)
In this equation y(t) represents the system output
vector, u(t) is the system input vector and e(t) is the
noise vector. In this heart rate application the noise
was assumed to be white, therefore it was not
included in the model. B(z) and A(z) are defined by
the following two equations
BzB
z
B
z
⋯B
z
(2)
A
z
1A
z
A
z
⋯A
z
(3)
Where z
is the backwards shift operator with
z
−1
y(t) = y(t − 1).
In this case the system input is the acceleration
signal, the output the heart rate signal. In MATLAB
the rivbjid function was used to calculate a model.
Both the order and the coefficients of the model
could vary. In the analysis different models with
different orders and time delays were calculated. All
the possible combinations for orders in numerator
and denominator going from one to three and for a
time delay from one to five samples were calculated.
This resulted in the calculation of 45 (3×3×5)
different models.
To choose which model fits best on the data the
Young identification criterion (YIC) was used. This
criterion is interesting since it combines a measure
of fit and parameter reliability. For a sample size N,
the YIC is defined as follows (Young, 2011)
1
(4)
Where
is the variance of the model errors,
the
variance of the data around the mean, the number
of parameters,
the value of uncertainty for the
i
th
parameter estimation and
the quadratic value
of the i
th
parameter. The best model is one with the
most negative YIC value, if this value is highly
positive it means the model is over-parameterised
(Young, 2011).
Following this procedure, a model was
calculated for the training, control and induced
anxiety session. From this model different
parameters could be calculated. The first group are
the error parameters, the second the individual
model parameters.
To calculate the error parameters the model that
was calculated based on the training data, was fit on
both the control and the induced anxiety data.
Consequently the error for both control and induced
anxiety session was calculated. This was done by
subtracting the actual heart rate data from the
simulated data. If it is assumed that the physical part
of heart rate is modelled in the training session and
that the metabolic part remains constant, then the
errors represent the mental part of the heart rate. In
this case a difference in anxiety level could become
visible through a difference between errors of the
control and induced anxiety session (Myrtek et al.,
2004). From this analysis nine different error
parameters were calculated.
The second group of parameters investigated the
properties of the models calculated on the three
sessions separately. These are the individual model
parameters. First the model orders were calculated.
This resulted in six parameters, being the order of
the numerator and denominator for training, control
and induced anxiety session. The time constants and
the steady state gains of the three models were also
calculated, which resulted in six additional
parameters. The time constant represents the time
that it takes for the model output the achieve 63,2%
of its final value as a response to a step input
(Lipták, 2006). The steady state gain is the ratio of
the change in output to the change in input when the
system has reached steady state as a response to a
step input (Lipták, 2006).
2.2.3 Statistical Analysis
The goal in the statistical analysis was to predict the
class to which a participant belonged based on the
parameter values calculated in the static and
dynamic analysis. This was done with the
multinomial logistic regression (MLR) model, which
is an extension of the binary logistic regression
model. It is used when the dependent variable exists
of different categories. One of these categories is
taken as the reference category. The probability of
an observation to belong to this reference category is
then compared with the probability to belong to one
of the other categories (Prabhakar et al., 2013). In a
first approach of the data, the classification was done
with the whole dataset both as training and as test
MonitoringtheMentalStatusofFootballPlayers
209
set. This resulted in the most optimal classification
results, but did not give an accurate representation of
the model performance. Therefore also a fivefold
cross-validation was performed. This means that
4/5th of the data was used as training set to create a
model and the remaining 1/5th was used as test set
to evaluate the performance of the model. This was
done five times with each time a different training
and test set. The division of the observations into
training or test set was randomly chosen, but it was
made sure that all the observations were exactly
once used as test set. For the five different
classifications each time the percentage with
correctly classified observations was calculated.
From these five results the mean was taken and this
percentage represents the overall performance of the
model.
3 RESULTS
The goal was to search for interesting parameters,
static or dynamic, that could be used as predictors in
the MLR model. This model could then be used to
predict to which class a participant belonged. In the
first section the results of the anxiety classification
are presented, in the second those of the penalty
performance classification.
3.1 Anxiety Classification
The result for the static analysis is presented in
Figure 1. The rows represent the class to which the
participant really belongs according to the reference,
the columns the class to which he was classified
according to the model or algorithm. This means that
the diagonal of the figure represents the correct
classified persons. The numbers represent the
number of participants that are classified in a certain
group. This classification however was done with
the whole dataset as both training and test set. This
resulted in the most optimal classification, but is not
an accurate reflection of the model performance.
Therefore also a five-fold cross-validation was
performed. This showed that 36,67% of the
participants were correctly classified using these
static parameters.
A similar analysis was done for the dynamic
parameters. There were however a total of 21
dynamic parameters, which is too much for an
effective model. Therefore first a MLR model was
calculated with all the 21 parameters. Afterwards
only the significant parameters were retained and a
new model with only these parameters was
calculated. The result of this model with only the
significant parameters can be seen in Figure 2. In
this case there were only three significant parameters
present. These were the mean of the anxiety error,
the time constant of the training session and the
steady state gain of the induced anxiety session.
After a five-fold cross-validation 47,11 % of the
participants were correctly classified.
Figure 1: Anxiety classification with static parameters.
Figure 2: Anxiety classification with the significant
dynamic parameters.
3.2 Penalty Classification
The result for the static analysis can be seen in
Figure 3. A five-fold cross-validation indicated that
26,22% of the participants were correctly classified.
For the dynamic analysis again only the
significant parameters were retained. The result can
be seen in Figure 4. In this case there are nine
significant parameters. These are the standard
deviation and the cumulative sum of the last part of
6
6
1
11
17
6
1
1
0
Algorithm
Reference
Anxiety Classification
Less anxiety No difference More anxiety
Less anxiety
No difference
More anxiety
3
1
0
14
23
7
1
0
0
Algorithm
Reference
A
nxiety Classification
Less anxiety No difference More anxiety
Less anxiety
No difference
More anxiety
icSPORTS2013-InternationalCongressonSportsScienceResearchandTechnologySupport
210
the control error, the model orders of both numerator
and denominator of the training session, the
denominator of the anxiety session and the time
constant and steady state gain of training and anxiety
session. The five-fold cross-validation indicated a
correct classification of 55,11%.
Figure 3: Penalty classification with static parameters.
Figure 4: Penalty classification with the significant
dynamic parameters.
4 DISCUSSION
This discussion consists of three sections. In the first
the classification results are discussed. The goal is to
investigate whether the hypothesis can be confirmed,
meaning that heart rate in combination with activity,
in this case represented by the acceleration signal,
can predict for both anxiety level and penalty
performance. In the second section a discussion of
the biological meaning of the different heart rate
parameters is presented. Finally some suggestions
for future work are given.
4.1 Classification Results
The most important observation is that generally the
classification percentages were not as high as hoped
for. This implicates that still some improvements are
necessary for the approach to be used in practice.
The best classification was obtained with nine
parameters from the dynamic analysis for the
penalty classification. Here a correct classification of
55,11 % was reached with a five-fold cross-
validation. An interesting solution to increase the
classification performance could be the use of a non-
linear classifier instead of the MLR model which is
linear. A possible method is by using support vector
machines. This is a statistical classification method
which was originally designed for binary
classification, but can be broadened for classification
into three groups as is the case in this experiment.
The method provides an optimal hyper plane that
separates the different classes (Bosch et al., 2013).
The advantage of this method is that it can use a
non-linear approach when necessary and therefore in
this situation can be more interesting for
classification than MLR. A second solution could be
to search for some additional parameters. For
example next to the time constant and steady state
gain there are still other dynamic properties of a
model such as the overshoot, rise time and settling
time. Furthermore also the results of the dynamic
approach can be improved by using a combination
of the R² and YIC value to select the best model.
Finally it needs to be mentioned that in general the
parameters of the dynamic approach performed
better than those of the static. Also the outcome of
the penalty in this experiment depends on the skill of
the player. The higher the skill level the more
consistent the player can be and this has an influence
on the penalty outcome. If someone is not consistent
then this can be the cause of the difference in
penalty outcome instead of the anxiety level.
Therefore it is suggested to focus in the future on the
dynamic approach and players with a similar, high
skill level.
It can be concluded that qualitatively the
hypothesis could be confirmed. This means that
heart rate in combination with activity could serve as
a predictor for both anxiety level and penalty
outcome. However, future research, with the
previous suggestions in mind, is necessary to
improve the quantitative results of the classification.
6
2
4
5
9
5
6
3
9
Algorithm
Reference
Penalty Classification
Less penalties No difference More penalties
Less penalties
No difference
More penalties
12
2
1
3
9
5
2
3
12
Algorithm
Reference
Penalty Classification
Less penalties No difference More penalties
Less penalties
No difference
More penalties
MonitoringtheMentalStatusofFootballPlayers
211
4.2 Biological Interpretation
Different parameters were used as predictors in the
MLR model for classification. It is not only the goal
to find these parameters, but also to search for an
explanation why exactly these parameters can make
the connection between heart rate and the mental
state. Especially the error parameters have attracted
researchers’ attention over the last years.
The effect of the mental activation on the
additional heart rate has already been investigated
(Myrtek et al., 2005). It is defined as the increase in
heart rate without a corresponding increase in
activity (Myrtek et al., 2005). Since the heart rate
dynamics due to activity were modelled, the error
reflects this additional heart rate (Jansen et al.,
2009). The hypothesis therefore is that when a
mental activation is present, the error should
increase. An important remark to keep in mind in
this context is that additional heart rate is also
influenced by parameters such as cardiac drift,
fatigue, etc. In this research this hypothesis is not
confirmed. Therefore it is suggested that future
research focuses more on this topic.
4.3 Future Work
The goal of this research was to predict the outcome
of a penalty shootout based on the measurements of
heart rate and activity. The underlying goal was to
find a physiological variable that could measure the
level of anxiety. The results have indicated that heart
rate has potential as predictor for anxiety. However,
the results are not yet good enough for practical
applications. In the previous section already some
possible improvements were listed. If these
suggestions are taken into account in future research
better results will become possible.
Furthermore it needs to be said that generally
research on the biological interpretation of model
parameters should increase. It is important to know
not only that some parameters could predict the
mental state, but also why this would be the case.
Finally, this research has only focused on
football. However, the need for a physiological
variable to measure anxiety or the mental state in
general is not restricted to this sport only. Future
research should test whether the algorithms
developed in this research are also applicable in
other sports. It is also important to broaden the
investigation further than only heart rate analysis. As
presented earlier also blood pressure and
biochemical variables such as epinephrine can
predict for anxiety. The downside of these variables
is that these cannot be used online which is an
important factor in the penalty shootout. However,
this is not equally important in every application.
Therefore these parameters should not be excluded
and more research should be dedicated to them.
5 CONCLUSIONS
As a general conclusion of this research it can be
said that the analysis of heart rate offers some
interesting perspectives for the future concerning the
measurement of anxiety. A follow-up study should
indicate whether better classification results can be
obtained when the different proposed adjustments
are implemented. Furthermore, more research should
be focused on finding a biological interpretation of
the parameters. Finally it is important to broaden the
research to other sports and other variables.
REFERENCES
Borg, G., (1998). Borg’s perceived exertion and pain
scales. Human kinetics publishers, Champaign. 104 p.
Bosch, P., López, J., Ramírez, H., Robotham, H., (2013).
Support vector machine under uncertainty: an
application for hydroacoustic classification of fish-
schools in Chile. Expert systems and applications, 40:
4029-4034.
Chen, M. J., Fan, X., Moe, S. T., (2002). Criterion-related
validity of the Borg ratings of perceived exertion scale
in healthy individuals: a meta-analysis. Journal of
sports sciences, 20: 873-899.
De Morree, H. M., Klein, C., Marcora, S. M. (2012).
Perception of effort reflects central motor command
during movement execution. Psychophysiology, 49:
1242-1253.
Frazier, S. K., Moser, D. K., Riegel, B., McKinley, S.,
Blakely, W., Kim, K. A., Garvin, B. J., (2002). Critical
care nurses’ assessment of patients’ anxiety: reliance
on physiological and behavioural parameters.
American journal of critical care, 11: 57-64.
Hoehn, T., Braune, S., Scheibe, G., Albus, M., (1997).
Physiological, biochemical and subjective parameters
in anxiety patients with panic disorder during stress
exposure. European archives of psychiatry and
clinical neuroscience, 247: 264-274.
Horikawa, M., Yagi, A., (2012). The relationships among
trait anxiety, state anxiety and the goal performance of
penalty shoot-out by university soccer players. Plos
one, 7(4).
Jansen, F., Van Der Krogt, J., Van Loon, K., Avezzù, V.,
Guarino, M., Quanten, S., Berckmans, D., (2009).
Online detection of an emotional response of a horse
during physical activity. The veterinary journal, 181,
38-42.
icSPORTS2013-InternationalCongressonSportsScienceResearchandTechnologySupport
212
Jordet, G., (2009). When superstars flop: public status and
choking under pressure in international soccer penalty
shootouts. Journal of applied sport psychology, 21,
125-130.
Knicker, A. J., Renshaw, I., Oldham, A. R., Cairns, S. P.,
(2011). Interactive processes link the multiple
symptoms of fatigue in sport competition. Sports
Medicine, 41: 307-328.
Lipták, B. G., (2006). Process control and optimization.
CRC Press, Florida, US. 67 p.
Marcora, S. M., (2009). Perception of effort during
exercise is independent of afferent feedback from
skeletal muscles, heart and lungs. Journal of applied
physiology, 106: 2060-2062.
Mateo, M., Blasco-Lafarga, C., Martínez-Navarro, I.,
Guzmán, J. F., Zabala, M., (2012). Heart rate
variability and pre-competitive anxiety in BMX
discipline. European journal of applied physiology,
112: 113-123.
Myrtek, M., Aschenbrenner, E., Brügner, G., (2005).
Emotions in everyday life: an ambulatory monitoring
study with female students. Biological Psychology, 68:
237-255.
Prabhakar, M., Prasad, Y. G., Desai, S., Thirupathi, M.,
Gopika, K., Rao, G. R., Venkateswarlu, B., (2013).
Hyperspectral remote sensing of yellow mosaic
severity and associated pigment losses in Vigna
mungo using multinomial logistic regression models.
Crop protection, 45: 132-140.
Taylor, C. J., Pedregal, D. J., Young, P. C., (2007).
System identification, time series analysis and
forecasting. The captain toolbox. Centre for
environmental systems and statistics, Lancaster,
England. 278 p.
Young, P. C., (2011). Recursive estimation and time-series
analysis. Springer, Berlin, Germany. 443 p.
Zamunér, A. R., Moreno, M. A., Camargo, T. M., Graetz,
J. P., Rebelo, A. C. S., Tamburús, N. Y., Da Silva, E.,
(2011). Assessment of subjective perceived exertion at
the anaerobic threshold with the Borg CR-10 scale.
Journal of sports and medicine, 10: 130-136.
Zhang, Z., Jia, J., Ding, R., (2012). Hierarchical least
squares based iterative estimation algorithm for
multivariable Box-Jenkins-like systems using the
auxiliary model. Applied mathematics and
computation, 218: 5580-5587.
MonitoringtheMentalStatusofFootballPlayers
213
