Machine learning approach to model sport training Edward Me ß _ zyk, Olgierd Unold ⇑ The Institute of Computer Engineering, Control and Robotics, Wroclaw University of Technology, Wyb. Wyspianskiego 27, 50-370 Wroclaw, Poland article info Article history: Available online 30 October 2010 Keywords: Machine learning Fuzzy logic Artificial immune system Data mining Sport training abstract The aim of this study was to use a machine learning approach combining fuzzy modeling with an immune algorithm to model sport training, in particular swimming. A proposed algorithm mines the available data and delivers the results in a form of a set of fuzzy rules ‘‘IF (fuzzy conditions) THEN (class)”. Fuzzy logic is a powerful method to cope with continuous data, to overcome problem of overlapping class definitions, and to improve the rule comprehensibility. Sport training is modeled at the level of microcycle and training unit by 12 independent attributes. The data was collected in two months (February–March 2008), among swimmers from swimming sections in Wrocław, Poland. The swimmers had minimum of 7 years of training and reached the II class level in swimming classification from 2005 to 2008. The goal of the performed experiments was to find the rules answering the question – how does the training unit influence swimmer’s feelings while being in water the next day? The fuzzy rules were inferred for two different scales of the class to be predicted. The effectiveness of the learned set of rules reached 68.66%. The performance, in terms of classification accuracy, of the proposed approach was com- pared with traditional classifier schemes. The accuracy of the result of compared methods is significantly lower than the accuracy of fuzzy rules obtained by a method presented in this study (paired t-test, P < 0.05). Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction In all sports, the key to success lies in a proper training planning and training execution on highest level. A well known problem and most popular in training preparation is finding an accurate stimu- lus in an appropriate time for each athlete. In efficiency sports solv- ing this problem is the key to success. The problem is even bigger when the training group is a combination of different types of ath- letes with diverse specialization. It is almost impossible to find studies that combine computer science and pint point this topic. Of course, a lot of works write in details about training monitoring and it is use in training planning (Maglischo, 2003; Friel, 2004). In Maglischo (2003) the author writes in details about blood testing procedures and procedures for monitoring heart rate with refer- ence to training monitoring and managing. In Friel (2004) the proper and effective methodical approach to making training logs, which are aimed to better experience usage in training preparing phase is described. For the sake of well-made training logs, the author proposes several parameters for monitoring athletes day disposition for training realization. Unfortunately none of this works, and even try to adopt machine learning algorithms to sup- port trainers decisions. Sport is a very wide discipline and only a few work about soft computing usage as a sport expert systems can be found (Bartlett, 2006; Rejman & Ochmann, 2007). Biomechanics are one of the first sport science disciplines that use goods provided by the artificial intelligence. Unfortunately, in efficiency sports like swimming, bio- mechanics are not so crucial for accuracy of training plan as proper training loads application. Determining training loads with the help of Artificial Neural Networks was proposed in Ryguła (2005). Exper- iments conducted and described in Ryguła work aimed to prove that training loads should be matched individually for each athlete. In Anderson, Hopkins, Roberts, and Pyne (2006) the authors evalu- ate the utility of incremental swimming as a method for monitoring seasonal changes in swimmers performance. A proposed training model takes into consideration the assump- tions, that training control should be based on a model and moni- toring (Ryguła, 2005; Anderson et al., 2006; Thaller, Tilp, & Sust, 2006). Model is based on simple training monitoring principles like heart rate measurements, regeneration time and basic training loads types (Maglischo, 2003; Friel, 2004). Other works which are in topic of training modeling and training evaluation via less dynamic parameters like anaerobic and aerobic thresholds, were also very helpful in building presented approach (Anderson et al., 2006; Saastamoinen, Ketola, & Turunen, 2004). Here, we show how machine learning approach combining a fuzzy modeling with an immune algorithm can be used to model sport training, in particular swimming training. The results 0747-5632/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.chb.2010.10.014 ⇑ Corresponding author. E-mail addresses: edward.mezyk@pwr.wroc.pl (E. Me ß _ zyk), olgierd.unold@pwr. wroc.pl (O. Unold). Computers in Human Behavior 27 (2011) 1499–1506 Contents lists available at ScienceDirect Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh