Analysing Cycling Sensors Data Through Ordinal Logistic Regression with Functional Covariates

Abstract

Abstract With the emergence of digital sensors in sports, all cyclists can now measure many parameters during their effort, such as speed, slope, altitude, heart rate or pedalling cadence. The present work studies the effect of these parameters on the average developed power, which is the best indicator of cyclist performance. For this, a cumulative logistic model for ordinal response with functional covariate is proposed. This model is shown to outperform competitors on a benchmark study, and its application on cyclist data confirms that pedalling cadence is a key performance indicator. However, maintaining a high cadence during long effort is a typical characteristic of high-level cyclists, which is something on which amateur cyclists can work to increase their performance.