Figure - available via license: CC BY
Content may be subject to copyright.
Source publication
This study aims to model training adaptation using Artificial Neural Network (ANN) geometric optimisation. Over 26 weeks, 38 swimmers recorded their training and recovery data on a web platform. Based on these data, ANN geometric optimisation was used to model and graphically separate adaptation from maladaptation (to training). Geometric Activity...
Contexts in source publication
Context 1
... the initial cohort of 38 swimmers, three dropped out due to time restrictions or changing swimming clubs, and one retired from swimming as illustrated in Figure 2. Thirteen swimmers met the prerequisites to the modelling (Table 4). The seven swimmers whose %PBT values were exclusively over 100%, had an average age of 15.3 ± 2.1 years, and their mean best FINA points value at the beginning of the study was 462 ± 78. ...Context 2
... the initial cohort of 38 swimmers, three dropped out due to time restrictions or changing swimming clubs, and one retired from swimming as illustrated in Figure 2. Thirteen swimmers met the prerequisites to the modelling (Table 4). The seven swimmers whose %PBT values were exclusively over 100%, had an average age of 15.3 ± 2.1 years, and their mean best FINA points value at the beginning of the study was 462 ± 78. ...Similar publications
We propose a framework for descriptively analyzing sets of partial orders based on the concept of depth functions. Despite intensive studies of depth functions in linear and metric spaces, there is very little discussion on depth functions for non-standard data types such as partial orders. We introduce an adaptation of the well-known simplicial de...
Citations
... However, deep learning models have been used for assessing key movement points and swimming techniques [8,9]. Studies that used these models in swimming prediction typically combined them with correlation and/or regression analysis for data interpretation [10][11][12], potentially limiting the accuracy of the artificial neural networks. ...
Explainable artificial intelligence (XAI) models with Shapley additive explanation (SHAP) values allows multidimensional representation of movement performance interpreted on both global and local levels in terms understandable to human intuition. We aimed to evaluate the swimming performance (World Aquatics points) predictability of a combination of demographic, training , anthropometric, and biomechanical variables (inputs) through XAI. Forty-seven swimmers (16 males), after completing a training questionnaire (background and duration) and anthropometric assessment, performed, in a randomised order, a 25 m front crawl and three countermovement jumps, at maximal intensity. The predicted World Aquatics points (516 ± 159; mean ± SD) were highly correlated (r 2 = 0.93) with the 529 ± 158 actual values. The duration of swimming training was the most important variable (95_SHAP), followed by the countermovement jump impulse (37_SHAP), both with a positive effect on performance. In contrast, a higher percentage of fat mass (21_SHAP) corresponded to lower World Aquatics points. Impulse, when interpreted together with dryland training duration and stroke rate, shows the positive effects of upper and lower limb power on swimming performance. Height should be interpreted together with arm span when exploring positive effects of anthropometric traits on swimming performance. The XAI modelling highlights the usefulness of specific training, technical and physical testing, and anthropometric factors for monitoring swimmers.
... A new surge in performance is possible due to better tracking, optimization, and (direct) control of the training process, which might be enhanced and automated in the near future using artificial intelligence. 80 Several variables of the model are difficult or impossible to measure directly in the daily training practice. For example, an indication of anaerobic power via the increase of blood lactate accumulation can only be obtained after exercise. ...
Swim coaches routinely monitor the performance and performance determining variables of their athletes to optimize training programs in an individual-specific manner with the ultimate aim to swim faster and win races. To interpret the collected data, they require a suitable, and practically useful, conceptual framework, which can be found in the power balance of swimming. For coaches a heuristic model for training and performance optimization based on the power balance has been proposed. We build on this model and illustrate how it can be employed in the training practice using data of an exemplary sub-elite swimmer (700 FINA points), which was collected continuously during two training sessions. Variables that are used in daily swim training practice, such as heart rate (HR), stroke count (SC), stroke rate (SR), and lap time (LT), were measured. By combining external load variables (e.g. LT) and internal load variables (e.g. HR) with technical variables (e.g. SR), the degree to which the swimmer complies with the training program can be determined. The measured values of these variables are presented using a traffic light feedback system indicating the degree of compliance. The traffic light system enables coaches to adjust the program if deemed necessary. It is thus shown how the model and commonly measured variables can assist swim coaches in the design and evaluation of training sessions in their pursuit of personal performance improvement leading to greater athletic success.
... Developing the methodology of the procedure aiming at defining standardized scientific-applied-metrological data processing can provide new knowledge useful in sports science, in general, or it can provide specific procedures in terms of improving the technology of sports training in swimming (Maglischo, 2003;Šiljeg, et al., 2017). Also, knowledge and understanding of the results obtained by measuring the level of blood lactate concentration in relation to the dynamics of its change in the recovery after maximum competitive loads will provide conditions for individualized implementation or correction of training work, which will enable a more effective process of training individualization and optimization as a current approach to the technology of training work with top-level athletes (Anderson, et al., 2006;Bonifazi, et al., 2000;Carrard, Kloucek, & Gojanovic, 2020;Costa, et al., 2013;Gulbin, Croser, Morley, & Weissensteiner, 2013;Lomax, 2012;Šiljeg, et al., 2017). ...
The goal of this study was to define the individual model characteristics of lactic acid removal after 200m breaststroke competitive load in a female swimmer in relation to different pool lengths (25m vs. 50m). The second goal was methodological and referred to the presentation of newly applied metrological procedures for the Individual Lactate Recovery Profile modeling. Six races from the competitive season 2021/22 were selected, in which the athlete achieved the most valuable results in relation to the FINA score. To establish the metabolic response of the organism to the competition effort, the method of determining the level of lactate concentration in capillary blood (La in mmol/L) was used. Differences between the mean values of variables were established using ANOVA. The polynomial curve equation function was used to create a blood lactate concentration in a function of recovery time model (La-trecovery). The ANOVA showed that there was no statistically significant difference between the monitored variables and the pool length function (p=.097). The maximum achieved blood lactate concentration in the acute race recovery phase was 13.17 ± 2.81 and 12.08 ± 1.80 mmol/L and the given concentration initially occurred in the time of 240.0 ± 85.6 s and 169.3 ± 79.9 s in the 25 and 50m pool, respectively. In relation to the time of complete passive recovery required to establish acidosis at the level of 2 mmol/L (25 and 50m pool) occurred in 1191.7 ± 481.3 s and 1326.7 ± 405.1 s, while the full index of intensity of blood lactate clearance was 135.7 ± 60.7 s/mmol/L and 124.0 ± 60.7 s/mmol/L for 25m and 50m pool, respectively. Although no statistically significant difference was found between the parameters of recovery in relation to pool length, the offered mathematical models enabled a practical individual approach to controlling the specific adaptation to training for achieving a higher competitive level performance.
... To date, there is a growing attraction of using ML models for modelling adaptations to training [33] as well as for predicting athletic performances [15,34,35]. That interest may be justified by a high predictive power of non-linear ML models in many applications, particularly when they are compared to FFMs or other univariate models in sports [15,35]. ...
The emergence of the first Fitness-Fatigue impulse responses models (FFMs) have allowed the sport science community to investigate relationships between the effects of training and performance. In the models, athletic performance is described by first order transfer functions which represent Fitness and Fatigue antagonistic responses to training. On this basis, the mathematical structure allows for a precise determination of optimal sequence of training doses that would enhance the greatest athletic performance, at a given time point. Despite several improvement of FFMs and still being widely used nowadays, their efficiency for describing as well as for predicting a sport performance remains mitigated. The main causes may be attributed to a simplification of physiological processes involved by exercise which the model relies on, as well as a univariate consideration of factors responsible for an athletic performance. In this context, machine-learning perspectives appear to be valuable for sport performance modelling purposes. Weaknesses of FFMs may be surpassed by embedding physiological representation of training effects into non-linear and multivariate learning algorithms. Thus, ensemble learning methods may benefit from a combination of individual responses based on physiological knowledge within supervised machine-learning algorithms for a better prediction of athletic performance.
In conclusion, the machine-learning approach is not an alternative to FFMs, but rather a way to take advantage of models based on physiological assumptions within powerful machine-learning models.
... The physiological adaptations involved by exercise being complex, some authors investigated the relationship between training and performance by using Artificial Neural Networks (ANN), non-linear machine learning models 35,36 . Despite low prediction errors reported (e.g. ...
This study aims to provide a transferable methodology in the context of sport performance modelling, with a special focus to the generalisation of models. Data were collected from seven elite Short track speed skaters over a three months training period. In order to account for training load accumulation over sessions, cumulative responses to training were modelled by impulse, serial and bi-exponential responses functions. The variable dose-response (DR) model was compared to elastic net (ENET), principal component regression (PCR) and random forest (RF) models, while using cross-validation within a time-series framework. ENET, PCR and RF models were fitted either individually (M_I) or on the whole group of athletes (M_G). Root mean square error criterion was used to assess performances of models. ENET and PCR models provided a significant greater generalisation ability than the DR model ( p = 0.018, p < 0.001, p = 0.004 and p < 0.001 for ENET_I , ENET_G, PCR_I and PCR_G, respectively). Only ENET_G and RF_G were significantly more accurate in prediction than DR (p < 0.001 and p < 0.012). In conclusion, ENET achieved greater generalisation and predictive accuracy performances. Thus, building and evaluating models within a generalisation enhancing procedure is a prerequisite for any predictive modelling.
... The AI, which employs the non-linear features of artificial neural networks (ANN), is an attractive tool for analysing swimming results. The accuracy of using ANN to predict the relative contributions of factors related to swimming performance has already been discussed in freestyle swimming (Heazlewood, 2006;De Jesus et al., 2019;Stanula et al., 2012), 200 and 400 m individual medley (Silva et al., 2007), 50 and 800 m freestyle swimming and backstroke start (Carrard et al., 2020;De Jesus et al., 2018). Swimming turn biomechanical studies of age-group swimmers have shown similarities between turning techniques Chainok et al., 2021), performance variable relationships (Blanksby et al., 1998(Blanksby et al., , 1996 and prediction of turning time using linear or multiple regression analysis (Blanksby et al., 1998Ling et al., 2004). ...
The purpose of the present study was to identify the performance determinant factors predicting 15-m backstroke-to-breaststroke turning performance using and comparing linear and tree-based machine-learning models. The temporal, kinematic, kinetic and hydrodynamic variables were collected from 18 age-group swimmers (12.08 ± 0.17 yrs) using 23 Qualisys cameras, two tri-axial underwater force plates and inverse dynamics approach. The best models were obtained: (i) with Lasso linear model of the leave-one-out cross-validation in open turn (MSE = 0.011; R² = 0.825) and in the somersault turn (MSE = 0.016; R² = 0.734); (ii) the Ridge of the leave-one-out cross-validation (MSE = 0.016; R² = 0.763) for the bucket turn; and (iii) the AdaBoost tree-based model of the leave-one-out cross-validation for the crossover turn (MSE = 0.016; R² = 0.644). Model’s selected features revealed that optimum turning performance was very similarly determined for the different techniques, with balanced contributions between turn-in and turn-out variables. As a result, the relevant feature’s contribution of each backstroke-to-breaststroke turning technique are specific; developing approaching speed in conjunction with proper gliding posture and pull-out strategy will result in improved turning performance, and may influence differently the development of specific training intervention programmes.
... Statistics and computer science show a great attraction over the last two decades in sports science and sports analytics, with high predictive power in particular for solving complex non-linear problems (Carrard, Kloucek, and Gojanovic 2020;Edelmann-Nusser, Hohmann, and Henneberg 2002;Mitchell, Rattray, Fowlie, et al. 2020). Machine-learning modelling approaches seek to approximate a function that maps the input (e.g. a combination of predictors) to the output (i.e. an athletic performance), lowering errors between predictions and observations in their way. ...
... More recently, two studies attempted to model responses to training using ANN (Carrard, Kloucek, and Gojanovic 2020;Mitchell, Rattray, Fowlie, et al. 2020). In the first, the authors focused on modelling adaptations and maladaptations to training using geometric optimisation. ...
The first models of training effects on athletic performance emerged with the work of Banister and Calvert through the so-called Fitness-Fatigue model (FFM). One major drawback of FFMs is that the features stem from a single source of data. That is not in line with the existing consensus about a multifactorial aspect of athletic performance. Hence, multivariate modelling approaches from statistics and machine-learning (ML) emerged.
A research issue arises from the quantification of training Loads (TL) in resistance training (RT) which lack of physiological evidence. In the first study, we provided a new method of TL quantification in RT based on physiological observations. To achieve that, we initially modelled the torque-velocity profiles of fifteen participants during an isokinetic leg extension task and assessed a set of physiological responses to various resistance exercises intensities. Each session was volume-equated according to the formulation of volume load (i.e. the product of the number of repetitions and the relative intensity).
Higher led to greater muscular fatigue described by neuromuscular impairments. Conversely, systemic and local pulmonary responses (measured through oxygen uptake) and metabolic changes (according to blood lactate concentrations) were more significant at low intensities, suggesting different contributions of metabolic pathways.
From these results, we provided a new index of TL based on the neuromuscu- lar impairments observed at exercise. We showed that to exponentially weight TL by the average rate decay of force development rate yielded better correla- tions with any of the significant physiological responses to exercise. In addition, information compressed within a principal component could be a valuable TL index.
In the second study, we provided a robust modelling methodology that relies on model generalisation. Using data from elite speed skaters, we compared a dose-response model to regularisation methods and machine-learning models.
Regularisation procedures provided the greatest performances in both generalisa- tion and accuracy. Also, we highlighted the pertinence of computing one model over the group of athletes instead of a model per athlete in a context of a small sample size.
Finally, ML approaches could be a way of improving FFMs through ensemble learning methods.
In the third study, we modelled acceleration-velocity directly from global posi- tioning system (GPS) measurements and attempted to predict the coefficients of the relationship between acceleration and velocity.
First, a baseline model was defined by time-series forecasting using game data only. Then, we proceeded to multivariate modelling using commercial features. A regularised linear regression and a long short term memory neural network were compared. Finally, we extracted features directly from raw GPS data and compared these features to the commercial ones for prediction purposes.
The results showed only slight differences between model accuracy, and no models significantly outperformed the baseline in the prediction task. Given the multi- factorial nature of athletic performance, using only GPS data for predicting such athletic performance criterion provided an acceptable accuracy. Using time-domain and frequency-domain features extracted from raw data led to similar performances compared to the commercial ones, despite being evidence-based. It suggests that raw data should be considered for future athletic performance and injury occurrence analysis.
Lastly, we developed an athlete management system for long-distance runners. This application provided an athlete monitoring module and a predictive module based on a physiological model of running performance.
A second development was realised under the SAP analytics cloud solution. Team management and automated dashboards were provided herein, in close collaboration with a professional Rugby team.
... Everyday a new application emerges. Recently, intelligent approaches for swimming data analysis are surfacing [21], [22] using intelligent machine learning approaches, e.g. Neural Networks (NN), Support Vector Machines (SVM) and Random Forests (RFo). ...
... The physiological adaptations involved by exercise being complex, some authors investigated the relationship between training and performance by using Artificial Neural Networks (ANN), non-linear machine learning models 35,36 . Despite low prediction errors reported (e.g. ...
This study aims to provide a transferable methodology in the context of sport performance modelling, with a special focus to the generalisation of models. Data were collected from seven elite Short track speed skaters over a three months training period. In order to account for training load accumulation over sessions, cumulative responses to training were modelled by impulse, serial and bi-exponential responses functions. The variable dose-response (DR) model was compared to elastic net (ENET), principal component regression (PCR) and random forest (RF) models, while using cross-validation within a time-series framework. ENET, PCR and RF models were fitted either individually (MI) or on the whole group of athletes (MG). Root mean square error criterion was used to assess performances of models. ENET and PCR models provided a significant greater generalisation ability than the DR model (p = 0.012, p < 0.001, p = 0.005 and p < 0.001 for ENETI, ENETG, PCRI and PCRG, respectively). Only ENETI, ENETG and RFI were significantly more accurate in prediction than DR (p = 0.020, p < 0.001 and p = 0.043, respectively). In conclusion, ENET achieved greater generalisation and predictive accuracy performances. Thus, building and evaluating models within a generalisation enhancing procedure is a prerequisite for any predictive modelling.
