fpsyg-11-02262 September 16, 2020 Time: 15:16 # 1
ORIGINAL RESEARCH
published: 18 September 2020
doi: 10.3389/fpsyg.2020.02262
Can Individual Movement
Characteristics Across Different
Throwing Disciplines Be Identified in
High-Performance Decathletes?
Fabian Horst1, Daniel Janssen2, Hendrik Beckmann1 and Wolfgang I. Schöllhorn1*
1 Department of Training and Movement Science, Institute of Sport Science, Johannes Gutenberg-University Mainz, Mainz,
Germany, 2 Gymnasium Dionysianum, Rheine, Germany
Although the individuality of whole-body movements has been suspected for years, the
scientific proof and systematic investigation that individuals possess unique movement
patterns did not manifest until the introduction of the criteria of uniqueness and
Edited by: persistence from the field of forensic science. Applying the criteria of uniqueness and
Ana Filipa Silva, persistence to the individuality of motor learning processes requires complex strategies
Polytechnic Institute of Maia, Portugal
due to the problem of persistence in the learning processes. One approach is to
Reviewed by:
Ana Paulo, examine the learning process of different movements. For this purpose, it is necessary
Université d’Orléans, France to differentiate between two components of movement patterns: the individual-specific
Howard N. Zelaznik, component and the discipline-specific component. To this end, a kinematic analysis
Purdue University, United States
*Correspondence: of the shot put, discus, and javelin throwing movements of seven high-performance
Wolfgang I. Schöllhorn decathletes during a qualification competition was conducted. In total, joint angle
schoellw@uni-mainz.de waveforms of 57 throws formed the basis for the recognition task of individual- and
Specialty section: discipline-specific throwing patterns using a support vector machine. The results reveal
This article was submitted to that the kinematic throwing patterns of the three disciplines could be distinguished
Movement Science and Sport across athletes with a prediction accuracy of up to 100% (57 of 57 throws). In
Psychology,
a section of the journal addition, athlete-specific throwing characteristics could also be identified across the
Frontiers in Psychology three disciplines. Prediction accuracies of up to 52.6% indicated that up to 10 out
Received: 13 May 2020 of 19 throws of a discipline could be assigned to the correct athletes, based on only
Accepted: 11 August 2020
Published: 18 September 2020 knowing these athletes from the kinematic throwing patterns in the other two disciplines.
Citation: The results further suggest that individual throwing characteristics across disciplines are
Horst F, Janssen D, Beckmann H more pronounced in shot put and discus throwing than in javelin throwing. Applications
and Schöllhorn WI (2020) Can for training and learning practice in sports and therapy are discussed. In summary, the
Individual Movement Characteristics
Across Different Throwing Disciplines chosen approach offers a broad field of application related to the search of individualized
Be Identified in High-Performance optimal movement solutions in sports.
Decathletes?
Front. Psychol. 11:2262. Keywords: motor learning, pattern recognition, high-performance sports, machine learning, support vector
doi: 10.3389/fpsyg.2020.02262 machine, individuality, transdisciplinary individuality
Frontiers in Psychology | www.frontiersin.org 1 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 2
Horst et al. Individuality Across Different Throwing Disciplines
INTRODUCTION suggested that before individuality can be assumed, one must
test the probability of uniqueness (indicating that no two persons
Most of us are familiar with the experience of identifying friends have identical characteristics) and the persistence/permanence
or colleagues by their walk (Cutting and Kozlowski, 1977), even of a physiological or behavioral characteristic (meaning that the
from a distance and with limited visibility (Stevenage et al., characteristic should be invariant with time).
1999). Practitioners in the field of sports science and physical The first steps toward such criteria took place in the
therapy often report the same experience of identifying individual analysis of everyday and sports movements and revealed the
athletes or patients based on their movement characteristics identification of individual people based on gait (Schöllhorn
(e.g., a characteristic forehand stroke in tennis or a unique et al., 1999, 2002; Nixon et al., 2006), running (Simon and
hand movement). Additionally, most of us have observed Schöllhorn, 1995), pole vaulting (Jaitner and Schöllhorn, 1995),
people mastering certain tasks easily, while struggling to become discus (Bauer and Schöllhorn, 1997), and javelin throwing
proficient in others. Both experiences serve as evidence of the (Schöllhorn and Bauer, 1998). The proposed approaches
individuality of human movements, though they may hold used self-organizing Kohonen maps, in combination with
various meanings and act epistemologically on different time cluster analysis, as early representatives for machine learning
scales (Newell et al., 2001). The tacit, universal acceptance classification in human movement science. The results indicated
of movement as a method of identifying individuals suggests the structural application of a statistical approach that is, similar
that most of us understand individualized movement, yet the to forensic proceedings, oriented toward a generic understanding
perfunctory nature of this acceptance has inhibited a deeper of probability. As follows, the probability of an event occurring
investigation of the concept’s essential aspects and consequences. is equal to the number of ways of achieving success relative to
In human movement science, anecdotal evidence has made the possible number of outcomes. For example, Schöllhorn and
claims of “individuality” for years (Bernstein, 1967; Marteniuk, Bauer (1998) recorded 10 javelin throws by a single athlete at
1974). Although the importance of individuality in sports different competitions over 5 years. Subsequently, the kinematic
training has been recognized since the origins of sports science patterns of these 10 throws were clustered together out of 51
(Matveev, 1970), the phenomenon has been mostly regarded as a kinematic patterns of throws from several other athletes. The
negligible side effect or as an exception in the search for universal probability of achieving this classification by chance is extremely
scientific laws (Harre, 1969, 2013; Matveev, 1970; Huber, 1977; low (<1 10−17× ). This outcome is far below the magnitude of
Schmidt and Young, 1991; Nitsch et al., 1997; Schnabel et al., common probabilities used, first, in the statistical model based
1997). In most cases, individuality appeared in the context on the work of Fisher and Mackenzie (1923) or Neyman and
of reliability studies that compared intra- and inter-individual Pearson (1928) and, second, in the magnitude of becoming
variance (Bates et al., 1983). These reliability studies led to the legally relevant.
standard requirement that an average of 10 to 25 movement Meanwhile, the uniqueness and persistence of individual
trials be conducted for each individual participant to achieve movement patterns could be validated for versatile whole-
an appropriate level of reliability or reproducibility (Bates et al., body movements such as walking (Horst et al., 2017b, 2019),
1983; DeVita and Bates, 1988; Gollhofer et al., 1990). The extent pedaling (Hug et al., 2019), basketball throwing (Schmidt,
to which the inter-individual variance distributions overlap to 2012), horse riding (Schöllhorn et al., 2006), or playing a
discriminate individuals from one another was not investigated. musical instrument (Albrecht et al., 2014). Up to this point,
In the past, the term individuality most often has been one might still be tempted to argue that a single movement
normatively applied in three ways: (1) when no classification pattern is optimal for an individual athlete. Initial doubts
criteria could be found (Brüggemann et al., 1991; Schöner et al., were raised, however, when it was remembered that none
1992; Button et al., 2000; Hecksteden et al., 2015; Barth et al., of the aforementioned studies could demonstrate identical
2019), (2) to explicitly circumvent “the difficulty of achieving repetitions of movement patterns, and thus strong indications
statistical significance” by creating smaller standard deviations were provided for the intuitively assumed (Bernstein, 1967)
and by describing several single cases (Davids et al., 1999; and previously biomechanically derived (Hatze, 1986) extremely
Button et al., 2000; Nuzzo, 2014), and (3) when individuality low probability of identical repetitions. Theoretically, however,
was predetermined in the form of case studies (Mendoza and the continuous fluctuations that were observed during the
Schöllhorn, 1990; Schöllhorn, 1993, Schöllhorn, 1998; Wallace proof of persistence could have been due to limitations in the
et al., 1994; Bauer and Schöllhorn, 1997; Button et al., 2006; biomechanical measurement resolutions or could have simply
Chow et al., 2006). Frequently, all three interpretations were used been random, unstructured noise. More detailed investigations
in combination and reflected a rougher approximation of the of fluctuations between repeated movement executions within
phenomenon than scientific evidence would suggest. individual persons surprisingly revealed strong evidence of fine
While movement and sports science still struggle to balance structures within a class of movement patterns. These finer
the demands of group-oriented science and individual athlete- structures showed a dependence on fluctuations in emotion
and patient-oriented practice, forensic science remains primarily (Janssen et al., 2008; Janssen, 2017), on fatigue (Jäger et al., 2003;
concerned with individual cases that must lend legal validity. Janssen et al., 2011), or on time (Bauer and Schöllhorn, 1997;
Therefore, the field of forensic science has developed specific Schöllhorn et al., 2002; Rein et al., 2010) with different time
methods and criteria for the identification of individuals (disjunct scales (Horst et al., 2016, 2017a). The individuality of movement
separation) (Kaye, 2010). In this context, Jain et al. (2006) patterns in connection with their fine structures thus indicate
Frontiers in Psychology | www.frontiersin.org 2 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 3
Horst et al. Individuality Across Different Throwing Disciplines
that individual movement patterns continuously change and TABLE 1 | Number of throwing trials per athlete and discipline.
adapt over time.
Athlete Age (years) Shot put Discus Javelin
In practice, the identification of athletes based on their
individual movement patterns and their corresponding fine A1 18–21 3 3 3
structure does not require individually tailored learning or A2 18–21 3 3 2
training methods. If we assume that individual differences A3 18–21 1 3 3
exist from birth, then theoretically, the same learning and A4 18–21 3 1 3
training content could have led to individually distinguishable A5 18–21 3 3 2
movement patterns at a later age. However, this awareness A6 18–21 3 3 3
is subject to the assumption that everyone responds in the A7 18–21 3 3 3
same way to intervention measures. To shed more light on Sum (disciplines) 19 19 19
these questions, further studies on individual learning behavior
should be conducted.
Indications for individual responses on a similar intervention in the kinematic throwing patterns of these disciplines using
came from physiological (Bouchard and Rankinen, 2001) and automatic classification by means of machine learning. Instead of
biomechanical studies (Cole et al., 1995; Schöllhorn et al., merely testing the individuality of athletes in a single throwing
2001, 2002). In the meantime, an increasing number of studies discipline, the present classifications are used to test whether
have observed phenomena that indicate the individuality of the kinematic throwing patterns will be assigned to the correct
adaptations and learning (Schöllhorn et al., 2006; Kostrubiec throwing discipline and whether the knowledge of, for example,
et al., 2012; Caballero et al., 2017). References to the advantages an individual athlete’s shot put movement patterns predicts
of learning with individual role models also began to question the individual’s discus or javelin movement patterns. For this
learning approaches based on average-oriented group role objective, high-performance athletes competing in a national
models (Brisson and Alain, 1996). Despite these initial signs, decathlon qualification competition were selected. This high
scientific evidence of individual learning processes according performance level served to guarantee sufficient stability for all
to the criteria of uniqueness and persistence is still missing three throwing movements. A competition was selected for this
(Fisher et al., 2018). Because of the normative nature of these study because it is a setting in which athletes often demonstrate
studies, a criteria-driven analysis, as proposed by Jain et al. their best performances, which we assume can increase the
(2006), is suggested. Applying the same criteria of uniqueness expression of the individuality in their movement patterns
and persistence to motor learning and adaptation processes (Schöllhorn et al., 2002).
requires, first, that each athlete/patient (e.g., in terms of changes
in movement patterns or performance outcomes) respond
differently to a particular intervention (uniqueness) and, second, MATERIALS AND METHODS
that individual responses to multiple interventions can be
repeatedly demonstrated (persistence). Seven right-handed, male decathletes (18.9 ± 0.4 years), who
While the first criterion could be tested indirectly via the were members of the German junior national team with at least
degree of learning progress each participant achieves, testing the 5 years of experience in the decathlon, were recorded during a
criterion of persistence is more complicated. The simplest way national decathlon qualifying competition. The final throwing
to prove the persistence of individual learning characteristics phases of 19 shot puts, 19 discus, and 19 javelin throws were
would be to allow athletes to learn the same skill several analyzed (Table 1). For right-handed athletes, the final throwing
times after wash-out phases. However, a limitation with this phases all begin when the left foot touches down and end when
approach has been raised by re-learning studies (Newell et al., the throwing object is released from the hand. Most of the
2001; Liu et al., 2003). Once a movement is acquired and increase in velocity of throwing object is produced during this
forgotten, it is re-learned more quickly (Malone et al., 2011). In phase (Hay, 1993; Bauersfeld and Schröter, 1998). The throwing
consequence, initial learning conditions cannot be reproduced performances ranged from 11.70 to 15.06 m (shot put), 33.66 to
exactly when a skill is re-learned several times, even with 43.74 m (discus), and 40.08 to 58.03 m (javelin).
adequate wash-out phases. This fact makes it almost impossible The recordings were taken using two high-frequency video
to compare the persistence of learning processes adequately. cameras (Weinberger MiniVis Eco-2; frequency: 200 fps;
Alternatively, individual characteristics of learning processes resolution: 1280 × 1024 pixel), which were positioned orthogonal
should be observable in the acquisition of different skills. to each other, one facing toward the flight direction of the
Therefore, finding approaches that can detach the individual- throwing object. A space of 3 3× 3 × 3 m was covered for the
specific characteristics from the task-specific characteristics in analysis using a calibration cube with 25 marker points. Due to
various movements would be helpful. Interestingly, previous the official competition rules, no marker points were allowed
studies on the uniqueness and persistence of movement patterns on the athletes. Data of both cameras were synchronized using
have only been carried out within a single movement task. an electric impulse transmitted from the master camera to the
This pilot study aims to analyze the three throwing disciplines slave camera during each throw. Neither the athletes nor the
in the decathlon (shot put, discus throw, and javelin throw) experienced digitizers were informed about the aim of the study.
and to search for athlete- and discipline-specific similarities The following anatomical body landmarks were digitized: the
Frontiers in Psychology | www.frontiersin.org 3 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 4
Horst et al. Individuality Across Different Throwing Disciplines
manubrium sterni, the left and right acromion, the epicondylus model was tested using the normalized kinematic patterns of
lateralis, the processus steyloidus ulnae, the spina illiaca anterior all discus throws.
superior, the trochanter major, the lateral end of the femur (knee), In the case of the discipline-classification, the SVMs were
the patella, the articulatio tibo fibulare talare, the calcaneus, the trained with the corresponding partitions of variable waveforms
phalanx distalis, and the hallux. of all athletes (except one) in all three throwing disciplines. The
The digitization of these points allowed the estimation of remaining waveforms of the one athlete were used to test the
three-dimensional joint angles of the right and left shoulder, performance of the SVM models for classification into one of the
elbow, hand, hip, knee, and ankle. All videos were manually three possible throwing disciplines.
digitized with SIMI Motion Software 5.0 (SIMI Reality Motion
Systems, Germany). Data were filtered using a recursive, second-
order Butterworth filter with a cutoff frequency of 14 Hz. All RESULTS
trials were digitized for an additional 10 video frames at both
phase boundary ends because of filter effects at the beginning and The results of the athlete-classification are shown in Table 2.
end of each signal. When the classification models were tested with data from
For javelin, the duration of the final throwing phase lasts the discus throws, the results showed the highest prediction
about 130 ms; for discus throwing, about 400 ms; and for shot accuracy of 52.6% for athletes when the SVM considered all
put, about 200 ms (Ballreich and Kuhlow, 1986; Ballreich et al., kinematic variables except the variables of the throwing arm. The
1989). Despite the variable duration of these final throwing lowest predictive accuracy of 21.1% was obtained when only the
phases, commonalities between all three throwing disciplines are kinematic data of the lower-body joint angles were considered.
assumed and used to economize training in combined events Similar results were found when the performance of SVM
(Hay, 1993). The trials were time normalized to 26 intervals to models was determined using the kinematic patterns of shot puts
compare the kinematic patterns of three disciplines. After time as test data. While the prediction accuracy is slightly lower when
normalization, the amplitudes were normalized over all trials and all variables are considered, the lowest prediction accuracy of
all athletes into the interval [0;1]. 31.6% is larger than for the discus split and is achieved when all
Time- and amplitude-normalized data formed the input variables and only the lower-body variables are used as test data.
vectors for the classifications using a support vector machine The lowest prediction accuracies are generally found when
(SVM) (Cortes and Vapnik, 1995). The classification of the the SVM models are tested based on javelin throwing patterns
kinematic throwing patterns based on SVM represents a (15.8–31.6%). When the SVM models for athlete-classification
supervised learning approach for pattern recognition in are trained on discus and shot put data, it seems to be more
data sets. The ability to distinguish kinematic throwing difficult to assign the movement patterns of javelin throwing to
patterns was investigated in multi-class classifications the individuals. Similar results can be observed in the pairwise
using a “one-vs.-all” algorithm. The L2-regularized, L2- cross-validations, which are also shown in Table 2.
loss, support-vector classification of the Liblinear Toolbox In Table 3, the results of the discipline-classification are
1.4.1 (Fan et al., 2008) was applied with a linear kernel listed with the same variable partitions as in Table 2. When
function within the software environment Scilab 6.0.2 all variables were included in the discipline-classification, the
(Scilab Enterprises, France). A grid search within the range respective disciplines could be predicted with an accuracy of
of C = 2−5, 2−4.75, , 215 was conducted to determine C 100%, based on the kinematic throwing patterns.. . .
experimentally before the training and testing of the SVM
models. An athlete-classification using a leave-discipline- DISCUSSION
out cross-validation and a discipline-classification using
a leave-athlete-out cross-validation were performed. This The results of this study reveal that the kinematic patterns
processing means that data from one discipline (in the case of the three throwing disciplines in the decathlon (shot
of the athlete-classification) or from one athlete (in the case put, discus throw, and javelin throw) could be distinguished
of the discipline-classification) were used either as training independently of the athlete with a prediction accuracy of up
or as test data during the cross-validation of the SVM to 100% (57 of 57 throws) using an automatic classification
models. A schematic overview of the entire approach with using machine learning (i.e., SVMs). In addition, prediction
data acquisition, processing, and classification is depicted in accuracies of up to 52.6% (10 of 19 throws) also indicate the
Figure 1. persistence of individual throwing characteristics of athletes
In the case of athlete-classification, the kinematic data of one across different throwing disciplines. The results further suggest
discipline were used in the cross-validation, either as training that individual throwing characteristics across disciplines are
or as test data (cf., athlete-classification in Figure 1). This more pronounced in shot put and discus throwing than in
use of data means that the classification model did not “see” javelin throwing. This finding demonstrates that the approach of
the throwing patterns of the athletes in the tested discipline classifying movement patterns using machine learning methods
during the training process. In the first cross-validation split, the allows for the identification of athlete- and discipline-specific
classification model was first trained to distinguish the athletes similarities in throwing patterns across different disciplines in
based on the normalized kinematic patterns of all shot puts high-performance athletes and suggests new ways to explore
and javelin throws. Then, the performance of the classification sports training in different disciplines.
Frontiers in Psychology | www.frontiersin.org 4 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 5
Horst et al. Individuality Across Different Throwing Disciplines
FIGURE 1 | Schematic overview of our approach to data acquisition, processing, and classification. (A) Seven right-handed athletes (A1–A7), who were members of
the German junior national team with at least 5 years of experience in the decathlon, were recorded during a national decathlon qualifying competition. The kinematic
analysis included all valid trials of the competition in the three throwing disciplines: S, shot put; D, discus throw; J, javelin throw. The final throwing phases of 19 shot
puts, 19 discus, and 19 javelin throws were analyzed using two orthogonally positioned high-frequency video cameras. After the digitization of 23 anatomical body
landmarks, the three-dimensional joint angles of the right and left shoulder, elbow, hand, hip, knee, and ankle were estimated. (B) Time- and amplitude-normalized
joint angle waveforms formed the input vectors for the classifications using a support vector machine (SVM). An athlete-classification using a leave-discipline-out
cross-validation and a discipline-classification using a leave-athlete-out cross-validation were performed. (C) The performance of the classification models was
assessed based on the prediction accuracy.
TABLE 2 | Prediction accuracy of the athlete-classification with leave-discipline-out cross-validation for different data partitions.
Test data Training Data Random baseline All variables All variables (without Only upper-body variables Only lower-body
throwing arm) (without throwing arm) variables
Discus Shot put and Javelin 14.3% (1/7 athletes) 42.1% (8/19 test trials) 52.6% (10/19 test trials) 47.4% (9/19 test trials) 21.1% (4/19 test trials)
Shot put Discus and Javelin 14.3% (1/7 athletes) 31.6% (6/19 test trials) 52.6% (10/19 test trials) 47.4% (9/19 test trials) 31.6% (6/19 test trials)
Javelin Discus and Shot put 14.3% (1/7 athletes) 15.8% (3/19 test trials) 21.6% (4/19 test trials) 31.6% (6/19 test trials) 31.6% (6/19 test trials)
Discus Shot put 14.3% (1/7 athletes) 36.8% (7/19 test trials) 52.6% (10/19 test trials) 52.6% (10/19 test trials) 42.1% (8/19 test trials)
Discus Javelin 14.3% (1/7 athletes) 26.3% (5/19 test trials) 36.8% (7/19 test trials) 21.1% (4/19 test trials) 26.8% (5/19 test trials)
Shot put Discus 14.3% (1/7 athletes) 47.4% (9/19 test trials) 57.9% (11/19 test trials) 47.4% (9/19 test trials) 31.6% (6/19 test trials)
Shot put Javelin 14.3% (1/7 athletes) 42.1% (8/19 test trials) 47.4% (9/19 test trials) 47.4% (9/19 test trials) 36.8% (7/19 test trials)
Javelin Discus 14.3% (1/7 athletes) 52.6% (10/19 test trials) 36.8% (7/19 test trials) 21.1% (4/19 test trials) 36.8% (7/19 test trials)
Javelin Shot put 14.3% (1/7 athletes) 10.5% (2/19 test trials) 15.8% (3/19 test trials) 26.3% (5/19 test trials) 26.3% (5/19 test trials)
In the following sections, we discuss the results in more detail. random baseline of 14.3%, provide a strong indication that
In the athlete-classification, the highest prediction accuracies individual movement signatures can be detected in different
by SVM models based on all variables except the throwing movements (e.g., different throwing disciplines). The present
arm variables mean that every second shot put kinematic findings reinforce previous studies that showed the uniqueness
pattern was correctly assigned to the corresponding individual and persistence of individual movement patterns within various
athlete when the SVM model was trained on the kinematic movements and support the call for a stronger focus on individual
throwing patterns of all athletes in javelin and discus throw. athletes or patients in sports and movement science (e.g.,
Prediction accuracies over ∼50%, which are well above the Horst et al., 2017b).
Frontiers in Psychology | www.frontiersin.org 5 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 6
Horst et al. Individuality Across Different Throwing Disciplines
TABLE 3 | Prediction accuracy of the discipline-classification with leave-athlete-out cross-validation for different data partitions.
Test data Training Data Random baseline All variables All variables (without Only upper-body variables Only lower-body
throwing arm) (without throwing arm) variables
A1 A2–A7 33.3% (1/3 disciplines) 100.0% (9/9 test trials) 100.0% (9/9 test trials) 100.0 (9/9 test trials) 100.0% (9/9 test trials)
A2 A1 and A2–A7 33.3% (1/3 disciplines) 100.0% (8/8 test trials) 100.0% (8/8 test trials) 100.0% (8/8 test trials) 100.0% (8/8 test trials)
A3 A1–A2 and A4–A7 33.3% (1/3 disciplines) 100.0% (7/7 test trials) 100.0% (7/7 test trials) 85.7% (6/7 test trials) 100.0% (7/7 test trials)
A4 A1–A3 and A5–A7 33.3% (1/3 disciplines) 100.0% (7/7 test trials) 100.0% (7/7 test trials) 100.0% (7/7 test trials) 100.0% (7/7 test trials)
A5 A1–A4 and A6–A7 33.3% (1/3 disciplines) 100.0% (8/8 test trials) 100.0% (8/8 test trials) 100.0% (8/8 test trials) 100.0% (8/8 test trials)
A6 A1–A5 and A7 33.3% (1/3 disciplines) 100.0% (9/9 test trials) 100.0% (9/9 test trials) 77.8% (7/9 test trials) 100.0% (9/9 test trials)
A7 A1–A6 33.3% (1/3 disciplines) 100.0% (9/9 test trials) 88.9% (8/9 test trials) 88.9% (8/9 test trials) 88.9% (8/9 test trials)
Note that the highest prediction accuracy is achieved using variables implies an automatic and differentiated recognition of
SVM models that consider all joint angle waveforms except the shot put, discus, and javelin throwing movement patterns. The
angles of the throwing arm. Comparatively lower prediction results provide promising evidence for the ability of pattern
accuracies using SVM models that take into account all joint recognition approaches using machine learning methods to
angle waveforms (including the ones of the throwing arm) might distinguish between different qualities of whole-body movements
be traced to a slightly reduced individuality and a predominant (Schöllhorn and Bauer, 1997; Schorer et al., 2007).
expression of the discipline specificity in the throwing-arm, Finally, some specificities of this pilot study should be
joint-angle waveforms. However, further research is needed to kept in mind. The chosen pattern recognition approach
determine whether this lower prediction accuracy is due to based on probabilities relative to the number of choices
the specificity of the disciplines or due to the variability in is distinguished from null-hypothesis-testing approaches. No
throwing arm movements. In this regard, a joint angle-wise claims for generalization are made. In addition, the demand
classification and determination of movement variability could for a relatively high level of performance in different sports
provide further clarification. disciplines limited the possibilities for empirical data collection
Higher prediction accuracies for SVM models based on the enormously. Some limitations arise from selecting decathletes
joint angles waveforms of the upper body without the throwing on their way from juvenile to adult competition classes as the
arm in shot put and discus throwing provide evidence for object for this pilot study. The athletes’ age suggests that some
increased individuality of the movement of the left arm, trunk, may not have completed puberty, and ongoing physical growth
and head in comparison to the waveforms of the lower-body joint could have an additional influence on the consistency of their
angles, which are more restricted by their contact to the ground. movement patterns. To what extent incomplete physical growth
Whether lower prediction accuracies of the SVM models based influences throwing patterns and throwing consistency requires
on lower-body joint angles are only due to the comparably coarse further research.
biomechanical data acquisition without anatomical markers or
due to the small geometric differences in the leg movements
cannot be resolved satisfactorily here. CONCLUSION
Considerably lower prediction accuracies of SVM models
for athlete-classification that were trained with the kinematic The results offer evidence for the possibility of automatic
patterns of shot put and discus throw and tested with javelin recognition of kinematic movement patterns originating from
throws are in line with findings of national (Kunz, 1980) different sports disciplines and confirm the assumption of a
and international (Pavlović and Idrizović, 2017) decathletes, strong and cross-disciplinary importance of individuality in
who showed a high correlation between performances in at least two of the throwing disciplines investigated. That
shot put and discus throwing, but no linear correlation certain individual movement characteristics can be identified in
with performance in javelin throwing. The finding that the the kinematic patterns of both shot put and discus throwing
individual throwing characteristics across the disciplines are is intriguing. This finding must be distinguished from the
more pronounced in shot put and discus throwing than in recognition of an individual athlete within a single discipline,
javelin throwing gives rise to the speculation about a more as shown for discus (Bauer and Schöllhorn, 1997) and javelin
individual coupling of the joint angles of the trunk and lower throwing (Schöllhorn and Bauer, 1998). An extension of
body with the left arm and head in shot put and discus this approach to the kinematic movement patterns in other
throwing. Future research is necessary to investigate whether sports disciplines such as the tennis serve, handball throw,
cross-disciplinary individual characteristics in shot put and or volleyball smash is reasonable. Exploring the respective
discus throwing also foster a positive transfer from training in proportion of individual characteristics in movement patterns
one discipline to the other. An analysis of individual muscle in more detail, even for dissimilar movement classes, will
activation signatures (Hug et al., 2019) during shot putting, be a challenge for future research. This exploration can be
discus, and javelin throwing could provide interesting insights compared with the search for analogies between different
in this context. biometric characteristics.
In discipline classification, a prediction accuracy of 100% A further criterion for individuality, which could be
for most cross-validation splits and combinations of considered summarized by homomorphism, could be added to the necessary
Frontiers in Psychology | www.frontiersin.org 6 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 7
Horst et al. Individuality Across Different Throwing Disciplines
criteria of uniqueness and persistence. Different from static movement patterns based on machine learning methods and
biometric measures such as fingerprints, facial characteristics, or the insights into the influencing factors indicated in this study
ear shapes, which are frequently directly related to static genetics, suggested that we are still at the beginning of understanding the
movement-based biometry is subject to dynamic changes and individuality of moving and learning human beings.
uncertain associations to the genome. While it is difficult
to find a common underlying basis for the biometrics of
finger, face, or ear apart from genetics, the comprehension of DATA AVAILABILITY STATEMENT
individual commonalities in different movements (e.g., throwing
disciplines) could provide access to the underlying individuality The raw data supporting the conclusions of this article will be
of central nervous physiology and structure. Future applications made available by the authors, without undue reservation, to any
of this approach could investigate the extent to which the central qualified researcher.
nervous system or the muscle physiology are modifiable beyond
an individual’s range.
Against this backdrop, the probability of finding a single
(time-independent) optimal movement pattern for an individual ETHICS STATEMENT
athlete is more than challenging. Instead, rethinking the
understanding of an optimal movement pattern is promising. Ethical review and approval was not required for the study
An extension of the term “optimal” by situation-optimal, as on human participants in accordance with the local legislation
the currently optimal solution for an individual athlete, may be and institutional requirements. Written informed consent for
initially tempting. However, an optimal solution would only serve participation was not required for this study in accordance with
as a theoretical model and could never be realistically achieved. the national legislation and the institutional requirements.
Because the motor system of an individual is constantly changing
and adapting, the model of a situation-optimal movement pattern
would also have to constantly change and adapt. Alternatively, AUTHOR CONTRIBUTIONS
the assumption of a situation-optimal model that is constantly
changing could be more advantageous for motor learning than All authors contributed to the article, critically revised the
for the pursuit of an insurmountable goal. manuscript, and approved the final version. WIS designed the
The study showed that an applied pattern recognition experiment. DJ, HB, and WIS conducted the acquisition and
approach based on a machine learning classification provides processing of the data. FH, DJ, and HB analyzed the data. FH, DJ,
an alternative and holistic approach for the analysis of and WIS designed the figure. FH and WIS interpreted the data
biomechanical movement data. This approach is closely and wrote the manuscript.
connected to a statistical method based on the original concept of
probabilities and may help to circumvent some of the limitations
connected with the Fisher and Mackenzie (1923) and Neyman FUNDING
and Pearson (1928) statistics.
Taken together, the findings of human movement science The study was funded by the German Federal Institute of Sport
regarding the uniqueness and persistence of individual Science (BISp; Grant No. 070613/08).
REFERENCES Bauersfeld, K. H., and Schröter, G. (1998). Grundlagen der Leichtathletik. Berlin:
Sportverlag.
Albrecht, S., Janssen, D., Quarz, E., Newell, K. M., and Schöllhorn, W. I. (2014). Bernstein, N. A. (1967). The Co-ordination and Regulation of Movements.
Individuality of movements in music - finger and body movements during New York, NY: Pergamon Press.
playing of the flute. Hum. Mov. Sci. 35, 131–144. doi: 10.1016/j.humov.2014. Bouchard, C., and Rankinen, T. (2001). Individual differences in response to
03.010 regular physical activity. Med. Sci. Sports Exerc. 33, S446–S451.
Ballreich, R., and Kuhlow, A. (1986). “Biomechanik des Kugelstoßes,” in Brisson, T. A., and Alain, C. (1996). Should common optimal movement patterns
Biomechanik der Leichtathletik, eds R. Ballreich, and A. Kuhlow (Stuttgart: be identified as the criterion to be achieved? J. Mot. Behav. 28, 211–223. doi:
Enke), 89–109. 10.1080/00222895.1996.9941746
Ballreich, R., Schöllhorn, W., and Menzel, H.-J. (1989). “Stoß- und Brüggemann, G. P., Gattermann, E., Göhner, U., Jander, H., Lampe, L., and Mester,
wurfdisziplinen,” in Biomechanik der Sportarten, ed. K. Willimczik (Reinbek: J. (1991). Individualisierung der Skitechnik – Grundlage für gesundes, sicheres
Rowohlt), 197–230. und freudvolles Skilaufen. Schriftreihe Dtsch. Skiverbandes 21, 25–48.
Barth, V., Käsbauer, H., Ferrauti, A., Kellmann, M., Pfeiffer, M., Hecksteden, A., Button, C., Davids, K., Bennett, S. J., and Taylor, M. A. (2000). Mechanical
et al. (2019). Individualized monitoring of muscle recovery in elite badminton. perturbation of the wrist during one-handed catching. Acta Psychol. 105, 9–30.
Front. Physiol. 10:778. doi: 10.3389/fphys.2019.00778 doi: 10.1016/s0001-6918(00)00044-5
Bates, B. T., Osternig, L. R., Sawhill, J. A., and James, S. L. (1983). An assessment of Button, C., Davids, K., and Schöllhorn, W. (2006). “Coordination profiling of
subject variability, subject-shoe interaction, and the evaluation of running shoes movement systems,” in Movement System Variability, eds K. Davids, S. Bennett,
using ground reaction force data. J. Biomech. 16, 181–191. doi: 10.1016/0021- and K. Newell, (Champaign, IL: Human Kinetics), 133–152.
9290(83)90125-2 Caballero, C., Moreno, F. J., Reina, R., Roldán, A., Coves, Á., and Barbado, D.
Bauer, H. U., and Schöllhorn, W. (1997). Self-organizing maps for the analysis of (2017). The role of motor variability in motor control and learning depends
complex movement patterns. Neural Process. Lett. 5, 193–199. doi: 10.1023/A: on the nature of the task and the individual’s capabilities. Eur. J. Hum. Mov. 38,
1009646811510 12–26.
Frontiers in Psychology | www.frontiersin.org 7 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 8
Horst et al. Individuality Across Different Throwing Disciplines
Chow, J. Y., Davids, K., Button, C., and Koh, M. (2006). Organization of motor Jaitner, T., and Schöllhorn, W. I. (1995). “Prozessorientierte Bewegungsanalyse am
system degrees of freedom during the soccer chip: an analysis of skilled Beispiel des Stabhochsprungs,” in Sport im Lebenslauf, eds D. Schmidtbleicher,
performance. Int. J. Sport Psychol. 37, 207–229. K. Bös, and A. Müller (Hamburg: Czwalina), 293–298.
Cole, G. K., Nigg, B. M., Fick, G. H., and Morlock, M. M. (1995). Internal loading Janssen, D. (2017). Analyse und Klassifikation der Effekte Unterschiedlicher
of the foot and ankle during impact in running. J. Appl. Biomech. 11, 25–46. Emotionsinduktionsmethoden auf das Menschliche Gangmuster Sowie den
doi: 10.1123/jab.11.1.25 Übergang Zwischen Gehen und Laufen. Hamburg: Kovac.
Cortes, C., and Vapnik, V. (1995). Support-vector networks. Mach. Learn. 20, Janssen, D., Schöllhorn, W. I., Lubienetzki, J., Fölling, K., Kokenge, H., and Davids,
273–297. doi: 10.1007/BF00994018 K. (2008). Recognition of emotions in gait patterns by means of artificial neural
Cutting, J. E., and Kozlowski, L. T. (1977). Recognizing friends by their walk: nets. J. Nonverbal Behav. 32, 79–92. doi: 10.1007/s10919-007-0045-3
gait perception without familiarity cues. Bull. Psychon. Soc. 9, 353–356. doi: Janssen, D., Schöllhorn, W. I., Newell, K. M., Jäger, J. M., Rost, F., and Vehof, K.
10.3758/BF03337021 (2011). Diagnosing fatigue in gait patterns by support vector machines and self-
Davids, K., Button, C., and Bennett, S. (1999). Modeling human motor systems organizing maps. Hum. Mov. Sci. 30, 966–975. doi: 10.1016/j.humov.2010.08.
in nonlinear dynamics: intentionality and discrete movement behaviors. 010
Nonlinear Dynamics Psychol. Life Sci. 3, 3–30. doi: 10.1023/A:1022429522099 Kaye, D. H. (2010). Probability, individualization, and uniqueness in forensic
DeVita, P., and Bates, B. T. (1988). Intraday reliability of ground reaction force science evidence. Brooklyn Law. Rev. 75, 1163–1185.
data. Hum. Mov. Sci. 7, 73–85. doi: 10.1016/0167-9457(88)90005-X Kostrubiec, V., Zanone, P.-G., Fuchs, A., and Kelso, J. A. S. (2012). Beyond
Fan, R. E., Chang, K.-W., Hsieh, C.-J., Wang, X.-R., and Lin, C.-J. (2008). Liblinear: the blank slate: routes to learning new coordination patterns depend on the
a library for large linear classification. J. Mach. Learn. Res. 9, 1871–1874. intrinsic dynamics of the learner-experimental evidence and theoretical model.
Fisher, A. J., Medaglia, J. D., and Jeronimus, B. F. (2018). Lack of group-to- Front. Hum. Neurosci. 6:222. doi: 10.3389/fnhum.2012.00222
individual generalizability is a threat to human subjects research. Proc. Natl. Kunz, H. (1980). Leistungsbesimmende Faktoren im Zehnkampf. Eine
Acad. Sci. U.S.A. 115, E6106–E6115. doi: 10.1073/pnas.1711978115 Laengsschnittstudie an Schweizer Spitzenathleten. Doctoral dissertation,
Fisher, R. A., and Mackenzie, W. A. (1923). Studies in crop variation: the manurial Universität Heidelberg, Heidelberg.
response of different potato varieties. J. Agric. Sci. 13, 311–320. doi: 10.1017/ Liu, Y.-T., Mayer-Kress, G., and Newell, K. M. (2003). Beyond curve fitting: a
s0021859600003592 dynamical systems account of exponential learning in a discrete timing task.
Gollhofer, A., Horstmann, G. A., Schmidtbleicher, D., and Schönthal, D. (1990). J. Mot. Behav. 35, 197–207. doi: 10.1080/00222890309602133
Reproducibility of electromyographic patterns in stretch-shortening type Malone, L. A., Vasudevan, E. V. L., and Bastian, A. J. (2011). Motor adaptation
contractions. Eur. J. Appl. Physiol. Occup. Physiol. 60, 7–14. doi: 10.1007/ training for faster relearning. J. Neurosci. 31, 15136–15143. doi: 10.1523/
BF00572179 JNEUROSCI.1367-11.2011
Harre, D. (1969). Trainingslehre. Berlin: Sportverlag. Marteniuk, R. G. (1974). Individual differences in motor performances and
Harre, D. (2013). Principles of Sports Training. Muskegon, MI: Ultimate Athlete learning. Exerc. Sport Sci. Rev. 2, 103–130.
Concepts. Matveev, L. P. (1970). Problemy izuèenija struktury trenirovki. Teor. Prak. Fiz. Kult.
Hatze, H. (1986). Motion variability–its definition, quantification, and origin. 33, 5–10.
J. Mot. Behav. 18, 5–16. doi: 10.1080/00222895.1986.10735368 Mendoza, L., and Schöllhorn, W. I. (1990). “Technical training in the field of high
Hay, J. G. (1993). The Biomechanics of Sports Techniques. Englewood Cliffs, NJ: performance athletes with a biomechanical feedback system,” in Proceedings
Prentice-Hall. of the Techniques in Athletics - The First International Conference, eds G. P.
Hecksteden, A., Kraushaar, J., Scharhag-Rosenberger, F., Theisen, D., Senn, S., Brüggemann, and J. K. Rühl, (Köln: Strauß), 412–419.
and Meyer, T. (2015). Individual response to exercise training - a statistical Newell, K. M., Liu, Y. T., and Mayer-Kress, G. (2001). Time scales in motor learning
perspective. J. Appl. Physiol. 118, 1450–1459. doi: 10.1152/japplphysiol.00714. and development. Psychol. Rev. 108, 57–82. doi: 10.1037/0033-295X.108.1.57
2014 Neyman, J., and Pearson, E. S. (1928). On the use and interpretation of certain test
Horst, F., Eekhoff, A., Newell, K. M., and Schöllhorn, W. I. (2017a). Intra- criteria for purposes of statistical inference: part I. Biometrika 20A, 175–240.
individual gait patterns across different time-scales as revealed by means of a doi: 10.1093/biomet/20A.1-2.175
supervised learning model using kernel-based discriminant regression. PLoS Nitsch, J. R., Neumaier, A., Mester, J., and Marées, H. D. (1997). Techniktraining:
One 12:e0179738. doi: 10.1371/journal.pone.0179738 Beiträge zu Einem Interdisziplinären Ansatz. Schorndorf: Hofmann.
Horst, F., Kramer, F., Schäfer, B., Eekhoff, A., Hegen, P., Nigg, B. M., et al. Nixon, M. S., Carter, J. N., Cunado, D., Huang, P. S., and Stevenage, S. V.
(2016). Daily changes of individual gait patterns identified by means of (2006). “Automatic gait recognition,” in Biometrics: Personal Identification in
support vector machines. Gait Posture 49, 309–314. doi: 10.1016/j.gaitpost.2016. Networked Society, eds A. K. Jain, R. Bolle, and S. Pankanti, (New York, NY:
07.073 Springer), 231–250.
Horst, F., Lapuschkin, S., Samek, W., Müller, K.-R., and Schöllhorn, W. I. (2019). Nuzzo, R. (2014). Scientific method: statistical errors. Nature 506, 150–152. doi:
Explaining the unique nature of individual gait patterns with deep learning. Sci. 10.1038/506150a
Rep. 9:2391. doi: 10.1038/s41598-019-38748-8 Pavlović, R., and Idrizović, K. (2017). Factor analysis of world record holders in
Horst, F., Mildner, M., and Schöllhorn, W. I. (2017b). One-year persistence of athletic decathlon. Sport Sci. 10, 109–116.
individual gait patterns identified in a follow-up study - A call for individualised Rein, R., Davids, K., and Button, C. (2010). Adaptive and phase transition
diagnose and therapy. Gait Posture 58, 476–480. doi: 10.1016/j.gaitpost.2017.09. behavior in performance of discrete multi-articular actions by degenerate
003 neurobiological systems. Exp. Brain Res. 201, 307–322. doi: 10.1007/s00221-
Huber, H. P. (1977). Single case analysis. Behav. Anal. Modifications 2, 1–15. 009-2040-x
Hug, F., Vogel, C., Tucker, K., Dorel, S., Deschamps, T., Le Carpentier, É., et al. Schmidt, A. (2012). Movement pattern recognition in basketball free-throw
(2019). Individuals have unique muscle activation signatures as revealed during shooting. Hum. Mov. Sci. 31, 360–382. doi: 10.1016/j.humov.2011.01.003
gait and pedaling. J. Appl. Physiol. 127, 1165–1174. doi: 10.1152/japplphysiol. Schmidt, R. A., and Young, D. E. (1991). Methodology for motor learning: a
01101.2018 paradigm for kinematic feedback. J. Mot. Behav. 23, 13–24. doi: 10.1080/
Jäger, J. M., Alichmann, M., and Schöllhorn, W. I. (2003). “Erkennung von 00222895.1991.9941590
Ermüdungszuständen anhand von Bodenreaktionskräften mittels Neuronaler Schnabel, G., Harre, D., and Borde, A. (1997). Trainingswissenschaft: Leistung -
Netze,” in Biologische Systeme – Mechanische Eigenschaften und Ihre Adaptation Training - Wettkampf. Berlin: Sportverlag.
bei Körperlicher Belastung, eds G. P. Brüggemann and G. Morey-Klapsing Schöllhorn, W. I. (1993). Biomechanische Einzelfallanalyse im Diskuswurf.
(Hamburg: Czwalina), 179–183. Frankfurt: Harri Deutsch.
Jain, A. K., Bolle, R., and Pankanti, S. (2006). “Introduction to biometrics,” in Schöllhorn, W. I. (1998). Systemdynamische Betrachtung komplexer
Biometrics: Personal Identification in Networked Society, eds A. K. Jain, R. Bolle, Bewegungsmuster im Lernprozess - Prozessorientierte Strukturierung
and S. Pankanti, (New York, NY: Springer), 1–41. doi: 10.1002/9780470997949. der Entwicklung eines Bewegungsablaufs mit Hilfe biomechanischer
ch1 Beschreibungsgrößen. Frankfurt: Peter Lang Verlag.
Frontiers in Psychology | www.frontiersin.org 8 September 2020 | Volume 11 | Article 2262
fpsyg-11-02262 September 16, 2020 Time: 15:16 # 9
Horst et al. Individuality Across Different Throwing Disciplines
Schöllhorn, W. I., and Bauer, H.-U. (1997). “Linear vs nonlinear classification of Schorer, J., Baker, J., Fath, F., and Jaitner, T. (2007). Identification of
complex time course patterns,” in Proceedings of the Second Annual Congress interindividual and intraindividual movement patterns in handball players of
of the European College of Sport Science, eds J. Bangsbo, B. Saltin, H. Bonde, varying expertise levels. J. Mot. Behav. 39, 409–421. doi: 10.3200/jmbr.39.5.
Y. Hellsten, B. Ibsen, M. Kjaer, et al. (Copenhagen: University of Copenhagen), 409-422
308–309. Simon, C., and Schöllhorn, W. I. (1995). “Verlaufsorientierte Strukturierung
Schöllhorn, W. I., and Bauer, H. U. (1998). “Identifying individual movement styles verschiedener Stützphasen des Sprintlaufs mit Hilfe der P- und
in high performance sports by means of self organizing Kohonen maps,” in S-Faktorenanalyse und Referenzfunktionen,” in Sport im Lebenslauf, eds
Proceedings of the XVI International Symposium on Biomechanics in Sports, eds D. Schmidtbleicher, K. Bös, and A. Müller (Hamburg: Czwalina), 299–302.
H. J. Riehle, and M. M. Vieten, (Konstanz: Universitätsverlag), 574–577. Stevenage, S. V., Nixon, M. S., and Vince, K. (1999). Visual analysis of gait as
Schöllhorn, W. I., Nigg, B. M., Stefanyshyn, D. J., and Liu, W. (2002). Identification a cue to identity. Appl. Cogn. Psychol. 13, 513–526. doi: 10.1002/(sici)1099-
of individual walking patterns using time discrete and time continuous data 0720(199912)13:6<513::aid-acp616>3.0.co;2-8
sets. Gait Posture 15, 180–186. doi: 10.1016/s0966-6362(01)00193-x Wallace, S. A., Stevenson, E., Spear, A., and Weeks, D. L. (1994). Scanning the
Schöllhorn, W. I., Peham, C., Licka, T., and Scheidl, M. (2006). A pattern dynamics of reaching and grasping movements. Hum. Mov. Sci. 13, 255–289.
recognition approach for the quantification of horse and rider interactions. doi: 10.1016/0167-9457(94)90040-X
Equine Vet. J. Suppl. 38, 400–405. doi: 10.1111/j.2042-3306.2006.tb05576.x
Schöllhorn, W. I., Röber, F., Jaitner, T., Hellstern, W., and Käubler, W. (2001). Conflict of Interest: The authors declare that the research was conducted in the
“Discrete and continuous effects of traditional and differential sprint training,” absence of any commercial or financial relationships that could be construed as a
in Proceedings of the Sixth Annual Congress of the European College of Sport potential conflict of interest.
Science, eds J. Mester, G. King, H. Strüder, E. Tsolakidis, and A. Osterburg,
(Köln: Sport und Buch Strauss), 331. Copyright © 2020 Horst, Janssen, Beckmann and Schöllhorn. This is an open-
Schöllhorn, W. I., Stefanysbyn, D. J., Nigg, B. M., and Liu, W. (1999). Recognition access article distributed under the terms of the Creative Commons Attribution
of individual walking patterns by means of artificial neural nets. Gait Posture License (CC BY). The use, distribution or reproduction in other forums is permitted,
10, 85–86. doi: 10.1016/S0966-6362(99)90454-X provided the original author(s) and the copyright owner(s) are credited and that the
Schöner, G., Zanone, P. G., and Kelso, J. A. (1992). Learning as change of original publication in this journal is cited, in accordance with accepted academic
coordination dynamics: theory and experiment. J. Mot. Behav. 24, 29–48. doi: practice. No use, distribution or reproduction is permitted which does not comply
10.1080/00222895.1992.9941599 with these terms.
Frontiers in Psychology | www.frontiersin.org 9 September 2020 | Volume 11 | Article 2262