INVITED COMMENTARY

International Journal of Sports Physiology and Performance, 2006;1:400-405

© 2006 Human Kinetics, Inc.

Using Mathematical Modeling

in Training Planning
Thierry Busso and Luc Thomas

This report aims to discuss the strengths and weaknesses of the application of
systems modeling to analyze the effects of training on performance. The simplifica-
tions inherent to the modeling approach are outlined to question the relevance of
the models to predict athletesʼ responses to training. These simplifications include
the selection of the variables assigned to the systemʼs input and output, the specifi-
cation of model structure, the collection of data to estimate the model parameters,
and the use of identified models and parameters to predict responses. Despite
the gain in insight to understand the effects of an intensification or reduction of
training, the existing models would not be accurate enough to make predictions
for a particular athlete in order to monitor his or her training.

Key Words: system identification, prediction, training schedule, performance

There is a growing interest in the application of systems theory to analyze the

responses to physical training. This interest was initiated by Banister and cowork-
ers,1 and their original model has elicited various extensions.2-5
Systems theory attempts to abstract a dynamic process into a mathematical
model. The system is an entity characterized by at least 1 input and 1 output related
by a mathematical law called transfer function (Figure 1). The output corresponds
to the systemʼs response to stimuli represented by the input. The transfer func-
tion characterizes the behavior of the system using parameters estimated from
observations. Working with models thus includes the selection of the variables
representing the systemʼs input and output, the identification of the model and its
parameters, and the prediction of the systemʼs behavior under given circumstances.6
This article attempts to outline some basic notions concerning the modeling of a
physiological process to question the usefulness of this approach for enhancing
sports performance.

Selection of Variables
Modeling training responses is a whole-body-based approach that considers perfor-
mance as the systemʼs output varying according to past training, which represents

The authors are with the Research Unit for Physiology and Physiopathology of Exercise and Handicap,
University of Saint-Etienne – PPEH, France.

400
 Predicting Performance 401

Figure 1 — Schematic representation of a system characterized by 1 output reacting to 1

input in accordance with a mathematical law, named transfer function, and a set of param-
eters estimated from observations.
Downloaded by ETSU on 09/17/16, Volume 1, Article Number 4

the systemʼs input. On one hand, the performance variable can be readily defined
according to the studied activity, and trials can be designed to collect data. Neverthe-
less, the subject has to perform the test frequently, whatever the period of training.
The need for a sufficient number of tests mimicking competition conditions is a
major difficulty that reduces the scope of model application in athletes. On the
other hand, the existing models are structured with a single input that represents
the total strain from training. This variable has to aggregate in a single entity the
total strain from various forms of exercise. A single input precludes accounting
for a possible specificity of the different types of exercise included in the training
program. This abstraction has required the creation of variables serving as the
systemʼs input. The existing methods of quantifying training have been reviewed
elsewhere.7,8 The problem has been solved in laboratory experiments because of
the standardization of training sessions, which allowed a simple assessment of the
variations in training loads.
These simple methods are, however, specific to these programs. When one
is using data from athletes, the problem becomes more arduous because of the
diversity of exercises. Training has been quantified in athletes from the training
volume weighted according to the intensity or type of exercise. In endurance activi-
ties, the weighting was done using an equation based on the average percentage of
cardiac reserve, giving rise to the so-called training impulse, or TRIMP.9 In other
applications, the weighting factors were arbitrarily fixed in accordance with the
intensity or the type of exercise. The stipulation of quantifying the training loads
in a single variable yielded thus to strong assumptions that also limited the scope
of model application. Multifactorial models could provide a better picture of the
effects of the various exercises included in the training program.1 Such a model
could consider several inputs acting on several factors that determine performance.
It would, however, be specific to a sports event and very demanding in terms of
data collection because of the number of variables and parameters to adjust.

Model and Parameter Identification

Controlled experiments and observational studies in athletes have provided data
used to specify the model of training responses and to estimate values for the
 402 Busso and Thomas

parameters. This situation, in which the input and output are known, presents the
problem of identification,6 that is, one wonders what systemʼs structure might
explain the observations (Figure 2). Owing to the complexity of the physiological
functions involved in the training adaptation, systems modeling can only aim to
abstract some particular features of the process. In this context, the modeling of
training responses has essentially focused on the dynamics of performance during
and after an intensification of training.8
A step in the analysis of model relevance is to assess the goodness of fit, which
indicates how well a model describes the observed data. The model proposed by
Banister et al1 assumed that the performance is the balance between negative and
positive antagonistic functions ascribed to fatigue and adaptation, respectively.
It was shown that this model significantly fit the training responses of subjects
enrolled in a controlled experiment10 and athletes involved in hammer throwing,11
weight lifting,12 running,13 and swimming.14,15 The adequacy of the fit supported
Downloaded by ETSU on 09/17/16, Volume 1, Article Number 4

the underlying theory of the model, which is that intensification of training would
result in performance decrement because the increase in fatigue would exceed the
gain in adaptation. A subsequent reduction of training would provide a dissipation
of fatigue more quickly than adaptation would, yielding to performance peak-
ing.10,16 This scheme was confirmed in a study of the responses of a group of elite
swimmers.14 These outcomes provided new insight to explain the performance
enhancement that results from reduced training during the final period before a
competition, known as the taper period.17
The statistical analysis is also used to compare the adequacy of competing
models in order to determine the best transfer function to describe an observed
phenomenon. For instance, a nonlinear model was built, assuming that the fatigue
induced by a training dose varies with the accumulation of training.4 This refinement
of the original model significantly improved the adequacy of the fit of the responses
of subjects who had trained on a cycle ergometer. A more accurate description than
that of earlier models supported the concept that a work session well endured during
habitual training could be more difficult to cope with when training is intensified.
Conversely, a more effective response could be restored with reduced training,

Figure 2 — Schematic representation of the procedure to specify 1 model for characterizing

a dynamical process. Assuming formulation of the transfer function, the model parameters
are estimated by fitting model output to actual data. The adequacy of the fit allows judging
the modelʼs relevance for describing the observation.
 Predicting Performance 403

which could contribute to performance peaking in addition to fatigue dissipation.4,17

This approach showed how a hypothesis about a physiological process can be
formulated into a mathematical model and tested using data from an intentional
experiment. A laboratory experiment in nonathletic subjects has the advantage of
enhancing the confidence in the parameter estimation and the discrimination in
the different models. The model parameters estimated in these conditions could,
however, not be representative of athletes in real situations.

Computer Simulations
Based on a relevant model and parameters, computer simulations can be used to
probe the systemʼs responses to solicitations (Figure 3). One can try to solve a
prediction or a control problem by seeking output for a given input or input for a
Downloaded by ETSU on 09/17/16, Volume 1, Article Number 4

given output, respectively, assuming that the systemʼs structure is known.6

In the area of modeling the responses to training, computer simulations have
essentially been performed in the interest of gaining insight into tapering effects. The
prediction from the existing models allowed us to go beyond the sole mathematical
formulation to define the implications for planning training. On one hand, using
the model from Banister et al,1 the parameters estimated in earlier studies provided
the simulation of the change in performance for given variations in training.18 This
highlighted the importance of the taper duration, which should be between approxi-
mately 2 and 4 weeks. In addition, a progressive reduction of training would be
more efficient than a step reduction, as indicated by the observations made during

Figure 3 — Schematic representation of the use of 1 model to predict the systemʼs behav-
ior under given circumstances. The specification of the model and the estimation of the
parameters from previous observation allow 2 types of theoretical analyses: the assessment
of the model output for a given input (prediction) or the research of the input that elicits a
given output (control).
 404 Busso and Thomas

an experiment in triathletes.19 On the other hand, the recent nonlinear model4

provided a more extensive analysis of the factors that could influence the optimal
taper characteristics including duration, extent, and form of training reduction.20
This theoretical analysis concluded that a period of intensified training before
the taper would enlarge the extent and duration of training reduction required to
maximize the performance. It was, however, based on model simulations using
parameters estimated from nonathletic subjects. Values for the model parameters
should be obtained from athletes in real situations in order to enhance the relevance
of the modelʼs predictions.
The ability of the model to indirectly estimate the response to a given train-
ing load might appear attractive to coaches for designing training programs. The
theoretical studies based on predictions from the model under given assumptions
should, however, be distinguished from using the model to control the training
program of a given individual. Using the model to monitor training would require
Downloaded by ETSU on 09/17/16, Volume 1, Article Number 4

a more precise assessment of the statistical confidence of the prediction.7,15 The

analysis of the adequacy of the performance fit allowed us to judge how 1 model
described past data, not how it could predict future data. The statistical confidence of
estimated model parameters could be diminished in athletes because of the difficul-
ties in data collection. In any case, it is likely that the expected accuracy between
model prediction and actual data will greatly suffer from the simplifications made
to aggregate total training strain in a single variable7 and, more generally, from
the abstraction of complex physiological processes into a very small number of
entities. If the intended purpose is to develop an instrument to help a coach pre-
scribe a training program for a particular athlete, new modeling strategies should
be considered in accordance with the specificity of the activity.7,15,21

Conclusion
The strong simplifications incurred when trying to model a dynamical process
would barely be compatible with the requirements of designing a training program.
The interest in the existing models could arise from the application of the insight
gained using models, rather than from a direct application of the models to a given
individual.

References
1. Banister EW, Calvert TW, Savage MV, Bach T. A systems model of training for athletic
performance. Aust J Sports Med. 1975;7:57-61.
2. Calvert TW, Banister EW, Savage MV, Bach T. A systems model of the effects of train-
ing on physical performance. IEEE Trans Syst Man Cybern. 1976;6:94-102.
3. Busso T, Denis C, Bonnefoy R, Geyssant A, Lacour JR. Modeling of adaptations
to physical training by using a recursive least squares algorithm. J Appl Physiol.
1997;82:1685-1693.
4. Busso T. Variable dose-response relationship between exercise training and performance.
Med Sci Sports Exerc. 2003;35:1188-1195.
5. Hellard P, Avalos M, Millet G, Lacoste L, Barale F, Chatard JC. Modeling the residual
effects and threshold saturation of training: a case study of Olympic swimmers. J
Strength Cond Res. 2005;19:67-75.
 Predicting Performance 405

6. Doucet P, Sloep P. Mathematical Modeling in the Life Sciences. New York, NY: Ellis
Horwood; 1992.
7. Taha T, Thomas SG. Systems modelling of the relationship between training and per-
formance. Sports Med. 2003;33:1061-1073.
8. Busso T. Modeling of adaptation and fatigue with overload training. In: Swan R, ed.
Trends in Exercise and Health Research. Hauppauge, NY: Nova Science Publishers;
2005:65-81.
9. Banister EW, Hamilton CL. Variations in iron status with fatigue modelled from train-
ing in female distance runners. Eur J Appl Physiol Occup Physiol. 1985;54:16-23.
10. Morton RH, Fitz-Clarke JR, Banister EW. Modeling human performance in running.
J Appl Physiol. 1990;69:1171-1177.
11. Busso T, Candau R, Lacour JR. Fatigue and fitness modelled from the effects of training
on performance. Eur J Appl Physiol Occup Physiol. 1994;69:50-54.
12. Busso T, Hakkinen K, Pakarinen A, Kauhanen H, Komi PV, Lacour JR. Hormonal
adaptations and modelled responses in elite weightlifters during 6 weeks of training.
Downloaded by ETSU on 09/17/16, Volume 1, Article Number 4

Eur J Appl Physiol Occup Physiol. 1992;64:381-386.

13. Wood RE, Hayter S, Rowbottom D, Stewart I.. Applying a mathematical model to
training adaptation in a distance runner. Eur J Appl Physiol. 2005;94:310-316.
14. Mujika I, Busso T, Lacoste L, Barale F, Geyssant A, Chatard JC. Modeled responses
to training and taper in competitive swimmers. Med Sci Sports Exerc. 1996;28:251-
258.
15. Hellard P, Avalos M, Lacoste L, Barale F, Chatard JC, Millet GP. Assessing the limita-
tions of the Banister model in monitoring training. J Sports Sci. 2006;24:509-520.
16. Banister EW. Modeling elite athletic performance. In: Green HJ, McDougal JD, Wenger
H, eds. Physiological Testing of Elite Athletes. Champaign, Ill: Human Kinetics;
1991:403-424.
17. Mujika I, Padilla S, Pyne D, Busso T. Physiological changes associated with the pre-
event taper in athletes. Sports Med. 2004;34:891-927.
18. Fitz-Clarke JR, Morton RH, Banister EW. Optimizing athletic performance by influence
curves. J Appl Physiol. 1991;71:1151-1158.
19. Banister EW, Carter JB, Zarkadas PC. Training theory and taper: validation in triathlon
athletes. Eur J Appl Physiol Occup Physiol. 1999;79:182-191.
20. Thomas L, Busso T. A theoretical study of taper characteristics to optimize performance.
Med Sci Sports Exerc. 2005;37:1615-1621.
21. Kinugasa T, Cerin E, Hooper S. Single-subject research designs and data analyses for
assessing elite athletesʼ conditioning. Sports Med. 2004;34:1035-1050.