suited for these analyses and corresponding data because they often account for multicollinearity and can model nonlinear relationships among large sets of variables. 16 This study will apply ML techniques to construct individual predictive models for professional soccer players to (1) examine
Tim Op De Beéck, Arne Jaspers, Michel S. Brink, Wouter G.P. Frencken, Filip Staes, Jesse J. Davis and Werner F. Helsen
Arne Jaspers, Tim Op De Beéck, Michel S. Brink, Wouter G.P. Frencken, Filip Staes, Jesse J. Davis and Werner F. Helsen
the same unit as the RPE value: an MAE of 1 means that, on average, the predicted RPE is 1 value below or above the reported RPE. Although an MAE of 0 is unrealistic, the goal is to minimize a model’s MAE. To construct predictive models, 2 standard machine learning techniques as well as 1 naive
Heidi R. Thornton, Jace A. Delaney, Grant M. Duthie, Brendan R. Scott, William J. Chivers, Colin E. Sanctuary and Ben J. Dascombe
To identify contributing factors to the incidence of illness for professional team-sport athletes, using training load (TL), self-reported illness, and well-being data.
Thirty-two professional rugby league players (26.0 ± 4.8 y, 99.1 ± 9.6 kg, 1.84 ± 0.06 m) were recruited from the same club. Players participated in prescribed training and responded to a series of questionnaires to determine the presence of self-reported illness and markers of well-being. Internal TL was determined using the session rating of perceived exertion. These data were collected over 29 wk, across the preparatory and competition macrocycles.
The predictive models developed recognized increases in internal TL (strain values of >2282 AU, weekly TL >2786 AU, and monotony >0.78 AU) to best predict when athletes are at increased risk of self-reported illness. In addition, a reduction in overall well-being (<7.25 AU) in the presence of increased internal TL, as previously stated, was highlighted as a contributor to self-reported-illness occurrence.
These results indicate that self-report data can be successfully used to provide a novel understanding of the interactions between competition-associated stressors experienced by professional team-sport athletes and their susceptibility to illness. This may help coaching staff more effectively monitor players during the season and potentially implement preventive measures to reduce the likelihood of illnesses occurring.
Youri Geurkink, Gilles Vandewiele, Maarten Lievens, Filip de Turck, Femke Ongenae, Stijn P.J. Matthys, Jan Boone and Jan G. Bourgois
included into the predictive model. Finally, individual deviations of several ELIs (total distance, total time, and number of sprints) compared with the group mean, based on historical training data, were derived. This way, the model could take differences in external workload for players compared with the
Alyssa Evans, Gavin Q. Collins, Parker G. Rosquist, Noelle J. Tuttle, Steven J. Morrin, James B. Tracy, A. Jake Merrell, William F. Christensen, David T. Fullwood, Anton E. Bowden and Matthew K. Seeley
to shoe; this is a manufacturing problem that has since been resolved. Given this shoe-to-shoe variability within the predictor variables, it did not make sense to create a single model to predict energy expenditure for all subjects. Rather, a predictive model was created for each shoe separately, by
Gil Rodas, Lourdes Osaba, David Arteta, Ricard Pruna, Dolors Fernández and Alejandro Lucia
, Tibshirani R . Regularization paths for generalized linear models via coordinate descent . J Stat Softw . 2010 ; 33 ( 1 ): 1 – 22 . PubMed ID: 20808728 doi: 10.18637/jss.v033.i01 20808728 23. Kuhn M . Building predictive models in R using the caret Package . J Stat Softw . 2018 ; 28 ( 5 ): 1 – 26
Lucie Péloquin, Pierre Gauthier, Gina Bravo, Guy Lacombe and Jean-Sébastien Billiard
The purposes of the present study were (a) to evaluate the test-retest reliability of the Price et al. (1988) 5-min walking field test, (b) to assess the validity of the test as an estimate of aerobic fitness, and (c) to derive a predictive model for estimating
Karyn Tappe, Ellen Tarves, Jayme Oltarzewski and Deirdra Frum
Predictive modeling for physical activity behavior has included many different psychological components, including planning, motivation, personality, and self-efficacy. However, habit formation in exercise maintenance has not been well explored and lacks reliable measurement tools. The current study explores novel survey questions that examine behavioral components of exercise habit, including frequency, environmental cuing, and temporal constancy of behavior. We then relate these concepts to an established psychological measure of habit, the Self-Report Habit Inventory (SRHI).
One hundred and seventy-four exercisers were surveyed at 2 private fitness clubs. A single questionnaire was administered that included the SRHI and the novel behavioral questions developed from habit formation concepts.
Habit formation was reported by many of the exercisers. Participants scoring higher on the SRHI also reported higher frequency of physical activity and a higher probability of environmental cuing. Exercise frequency did not correlate well with environmental cuing.
Habit formation appears relevant to the physical activity patterns of many regular exercisers. However, wide variation in response styles was evident suggesting further development and exploration of the novel questionnaire is warranted. The ultimate goals are to include habit in predictive models of physical activity, and then to inform interventions to increase exercise adherence.
Glenn M. Street and Robert W. Gregory
While the scientific literature has confirmed the importance of high maximal aerobic power to successful cross-country skiing performance, the same cannot be said of skiing technique or gliding characteristics of skis. The purpose of this study was to determine whether glide speed was related to Olympic race performance. Male competitors in the 50-km freestyle event were videotaped during the 1992 Winter Olympic Games. Glide speeds of the entire field were measured through a 20-m flat section at the bottom of a 150-m, 12° downhill. A significant correlation (r = -.73) was found between finish time and glide speed, showing that the more successful competitors tended to have faster glide speeds through this section of the course. A predictive model of glide speed suggested that the faster glide speeds were due primarily to differences in friction. There was little evidence to suggest that differences in air drag, body mass, or initial speed accounted for the major differences in glide speeds.
James J. Dowling and Lydia Vamos
Subjects performed maximum vertical jumps on a force platform to reveal whether resulting force-time curves could identify characteristics of good performances. Instantaneous power-time curves were also derived from the force-time curves. Eighteen temporal and kinetic variables were calculated from the force- and power-time curves and were compared with the takeoff velocities and maximum heights via correlation and multiple regression. The large variability in the patterns of force application between the subjects made it difficult to identify important characteristics of a good performance. Maximum positive power was found to be an excellent single predictor of height, but the best three-predictor model, not including maximum power, could only explain 66.2% of the height variance. A high maximum force (> 2 body weights) was found to be necessary but not sufficient for a good performance. Some subjects had low jumps in spite of generating high peak forces, which indicated that the pattern of force application was more important than strength.