Modelagem estatística multivariada da relação entre treinamento e performance no ciclismo

Abstract

Cycling is a growing sport with all over the world. The use of sensors to measure several variables during practice makes it attractive for analysis and research, in addition to providing input for the application of statistical inference methodologies appropriate for big data. This thesis aims to identify the influence of training intensity variables on the variation of performance parameters, for example critical power (𝐶𝑃), using multivariate modeling. Unlike other sports whose performance quantification can be simplified, cycling dynamics requires more sophisticated methodologies to measure an athlete's potential. The power­ duration models studied in this thesis provide parameters that represents human energy systems. The database comes from the Golden Cheetah Open Data Project repository, which contains sports activities of Golden Cheetah software users. The multivariate modeling is built upon neural networks models with estimation performed by stochastic gradient methods, due to the large volume of data. The relationship between training and the different performance variables is modeled simultaneously.