Longitudinal Study of Effects of Elevation Training on Cycling Performance: Cluster Analysis and Linear Mixed Effects Model

• Autor(a) principal: Ferreira, Martina Lopes
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/152349
• topic: Cycling performance; Power output; Cluster analysis; Repeated measures; Linear mixed-effects model; Domínio/Área Científica::Ciências Naturais::Matemáticas

In cycling, for cyclists to be able to keep track of their evolution, performance can be measured using tests of maximal effort or, when those cannot be performed, parameters obtained on submaximal efforts, such as power output. Usually, training programs with specific elevation gains are prescribed in order to increase training load. With that in mind, the purpose of this study was, through data analysis techniques, namely cluster analysis and linear mixed-effects models, to compare different elevation gain profiles and determine whether they had a significant effect on performance evolution or not. To accomplish that goal, a database, available on the internet with name GoldenCheetah Open Data was utilized. It contained 2681 athletes, to whom inclusion and exclusion criteria were applied, resulting in a final 1308 cyclists included. Inclusion criteria were being 10 to 90 years old and having at least cycling 20 activities recorded. Exclusion criteria were all activities have missing values and all activities had outlier observations. For descriptive analysis, summarytools from R software was used. Cyclists were 42 years old, on average, and, from the 1308 athletes, 32 were women. In cluster analysis, cluster, imputeTS, factoextra and clValid libraries from R were used to apply four clustering algorithms in order to test and compare with stability measures. Through this analysis we concluded K-means algorithm was the one that performed better, which is why it was analyzed, and two clusters were obtained as a result: the cluster with high elevation gains and the cluster with low elevation gains. Significance tests showed a significant difference on age between clusters, but not on the proportion of female gender. As for training parameters, all were significantly different between clusters, including the ones related to power output. At last, with rstatix and nlme libraries from R, a linear mixed-effects model was applied using all response variables. However, after a moderate correlation was found among some variables, using corrplot R library, a new model with age, gender, number of activity, elevation gain, average speed and average heart rate was applied. The last linear mixed-effects model for average power revealed a significant negative influence of female gender and a significant positive influence of elevation gain, number of activity, average speed, average heart rate and the interaction between elevation gain and the number of the activity. The linear mixed-effects model for peak power on both 5 and 30 minute efforts evinced a significant negative influence of both age and female gender and a significant positive influence of elevation gain, number of activity, average speed, average heart rate and the interaction between elevation gain and the number of the activity. It was possible to conclude that both cluster analysis and repeated measures analysis were effective establishing a relationship between including vertical climbing on training programs and better performance improvement.