Informatica 41 (2017) 517–518 517
Computational Intelligence Algorithms for the Development of an Artificial
Sport Trainer
Iztok Fister Jr.
University of Maribor, Faculty of Electrical Engineering and Computer Science, Koroška 34, 2000 Maribor
E-mail: iztok.fister1@um.si
Thesis summary
Keywords: computational intelligence, nature-inspired algorithms, sport training
Received: October 12, 2017
This paper presents a short summary of doctoral thesis that proposes the use of computational intelligence
algorithms for the development of an Artificial Sport Trainer.
Povzetek: Članek predstavlja kratko vsebino doktorske disertacije, ki predlaga uporabo algoritmov
računske inteligence za razvoj umetnega športnega trenerja.
1 Introduction
Planning the proper sport training sessions for athletes is
a very hard problem for sports trainers. With the rising
computational power on the one hand, and emerging data
warehouses on the other, new algorithms for discovering
knowledge from these data have emerged. An overview of
literature in this domain showed that there is still a lack
of algorithms for knowledge enrichment from data that are
based explicitly on computational intelligence [1]. Inte-
restingly, there were some solutions that applied artificial
neural networks for controlling the sports training sessi-
ons (e.g. [5, 6]) and even fuzzy logic [7]. However, al-
most none of the studies suggested the use of population-
based nature-inspired algorithms [4] (e.g., evolutionary al-
gorithms, or swarm intelligence algorithms) for these tasks.
Contrary, the thesis [3] proposes a concept of an intelli-
gent system called Artificial Sport Trainer (AST) for the
training purposes of athletes. AST is based on stochas-
tic population-based nature-inspired algorithms, designed
to cover all phases of the sports training, which are also
in the domain of a real sport trainer: planning, realization,
control and evaluation.
The thesis is divided into two parts. The first, theore-
tical part, presents the fundamentals of computational in-
telligence, basics of sports training and the architecture of
the AST. The second, experimental part, presents two ap-
plications of the AST. The former is devoted to planning
sports training sessions based on existing sports activities,
e.g. comprehensive performance study of six different sto-
chastic, nature-inspired population-based algorithms (Bat
Algorithm (BA), Differential Evolution (DE), Firefly Al-
gorithm (FA), Hybrid Bat Algorithm (HBA), self-adaptive
Differential Evolution (jDE) and Particle Swarm Optimi-
zation (PSO)). These algorithms were tested on three real
datasets, i.e. professional cyclist, amateur cyclist and semi-
professional runner. The latter proposes a solution for As-
sociation Rule Mining (ARM) based on BA (the so-called
BatMiner), which is applied to real datasets for finding cy-
clist’s characteristics during the sports training.
2 Artificial sport trainer
The main architecture of the AST [2] covers the following
phases of the sport training:
– Planning: the most important phase of the sport trai-
ning, that consists of:
– long-term planning (so-called strategy) and
– short-term planning (so-called tactics).
– Realization: this phase captures the realization of the
sports training session.
– Control: realization of the sports sessions is typically
controlled by wearable devices, such as sports wat-
ches or smart-phones.
– Evaluation: after the conducted training plan, the ex-
pected form or abilities of an athlete are evaluated.
3 Experiments and results
In order to confirm that the AST can be used in practice,
we have conducted a comprehensive experimental work
that includes mentioned six different algorithms (i.e., BA,
DE, FA, HBA, jDE, PSO) on three real datasets obtai-
ned by different kinds of athletes (i.e., professional, semi-
professional and amateur) in two sports (i.e. cycling, run-
ning). Additionally, we have also studied the influence
of clusters that was obtained by k-means clustering. We
have used the following numbers of clusters: 5, 8, 10, 12,
518 Informatica 41 (2017) 517–518 I. Fister Jr.
15, and 18. The second application was tested on a real
cyclist’s dataset and was compared to the Hybrid Binary
Cuckoo Search for ARM. Resulting plans of the first appli-
cation have then been compared to the plans, created by a
real sport trainer. Comparison showed that the AST can be
used for planning sport trainings sessions, according to the
TRIMP indicator with confidence of 0.1. The results of the
second application showed that a BatMiner is an appropri-
ate algorithm for finding characteristics of athletes during
the sports training.
4 Conclusion
Main findings of the thesis [3] are: (1) A new research area
is proposed, i.e., use of computational intelligence algo-
rithms in the sport area, (2) The concept of an Artificial
Sport Trainer encompasses various algorithms of com-
putational intelligence in sport, (3) New population-based
nature-inspired algorithms for planning sport training ses-
sions are developed and validated on the real data obtained
by two cyclists and one runner, (4) An easy metric for com-
paring AST’s and real trainer’s session plans is proposed
and (5) The BatMiner algorithm for mining characteristics
of athletes during the sports training sessions is built.
References
[1] Andries P Engelbrecht. Computational intelligence:
an introduction. John Wiley & Sons, 2007.
[2] Iztok Fister, Karin Ljubič, Ponnuthurai Nagaratnam
Suganthan, and Matjaž Perc. Computational intelli-
gence in sports: challenges and opportunities within a
new research domain. Applied Mathematics and Com-
putation, 262:178–186, 2015.
[3] Iztok Fister Jr. Algoritmi računske inteligence za raz-
voj umetnega športnega trenerja. Doctoral thesis, Uni-
versity of Maribor, Slovenia, 2017.
[4] Iztok Fister Jr, Xin-She Yang, Iztok Fister, Janez
Brest, and Dušan Fister. A brief review of nature-
inspired algorithms for optimization. arXiv preprint
arXiv:1307.4186, 2013.
[5] Hristo Novatchkov and Arnold Baca. Machine learning
methods for the automatic evaluation of exercises on
sensor-equipped weight training machines. Procedia
Engineering, 34:562–567, 2012.
[6] Hristo Novatchkov and Arnold Baca. Artificial intel-
ligence in sports on the example of weight training.
Journal of sports science & medicine, 12(1):27, 2013.
[7] Hristo Novatchkov and Arnold Baca. Fuzzy logic in
sports: a review and an illustrative case study in the
field of strength training. International Journal of
Computer Applications, 71(6), 2013.