Training load responses modelling and model generalisation in elite sports

Frank Imbach,Thibaut Meline,Robin Candau,Romain Chailan,Stephane Perrey

doi:10.1038/s41598-022-05392-8

Frank Imbach, Thibaut Meline + Show 3 more

Open Access

https://doi.org/10.1038/s41598-022-05392-8

Copy DOI

Abstract

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.

Highlights

Mixed model analysis showed that both ENET and principal component regression (PCR) models lowered the differences in terms of prediction errors between the training and evaluation data set
We provided a transferable modelling methodology which relies on the evaluation of models generalisation ability in a context of sport performance modelling
The mathematical variable dose-response model along Elastic net, principal component regression and random forest models were cross-validated within a time series framework

Summary

Introduction

A more statistical approach was used to investigate the effects of training load on performance by using principal component analysis and linear mixed models on different time frames[12] Such models infer parameters from all available data (i.e. combining subjects instead of by-subject model) but allow parameters to vary regarding the heterogeneity between athletes. While aforementioned studies aimed to describe the training load - performance relationships by estimating model parameters and by testing the model on a single data set, generalisation of models cannot be ensured. This challenges their usefulness in a predictive application. Because generalisation ability is not systemically appraised, practical and physiological interpretations drawn from some models may be incorrect and at least should be taken with caution

Objectives

Methods

Results

Discussion

Conclusion