Real-Time Power Performance Prediction in Tour de France

Pro-cycling team cyclist assignment for an upcoming race

Professional bicycle racing is a popular sport that has attracted significant attention in recent years. The evolution and ubiquitous use of sensors allow cyclists to measure many metrics including power, heart rate, speed, cadence, and more in training and racing. In this paper we explore for the first time assignment of a subset of a team's cyclists to an upcoming race. We introduce RaceFit, a model that recommends, based on recent workouts and past assignments, cyclists for participation in an upcoming race. RaceFit consists of binary classifiers that are trained on pairs of a cyclist and a race, described by their relevant properties (features) such as the cyclist's demographic properties, as well as features extracted from his workout data from recent weeks; as well additional properties of the race, such as its distance, elevation gain, and more. Two main approaches are introduced in recommending on each stage in a race and aggregate from it to the race, or on the entire race. The model training is based on binary label which represent participation of cyclist in a race (or in a stage) in past events. We evaluated RaceFit rigorously on a large dataset of three pro-cycling teams' cyclists and race data achieving up to 80% precision@i. The first experiment had shown that using TP or STRAVA data performs the same. Then the best-performing parameters of the framework are using 5 weeks time window, imputation was effective, and the CatBoost classifier performed best. However, the model with any of the parameters performed always better than the baselines, in which the cyclists are assigned based on their popularity in historical data. Additionally, we present the top-ranked predictive features.

Predicting the next Pogačar: a data analytical approach to detect young professional cycling talents

The importance of young athletes in the field of professional cycling has sky-rocketed during the past years. Nevertheless, the early talent identification of these riders largely remains a subjective assessment. Therefore, an analytical system which automatically detects talented riders based on their freely available youth results should be installed. However, such a system cannot be copied directly from related fields, as large distinctions are observed between cycling and other sports. The aim of this paper is to develop such a data analytical system, which leverages the unique features of each race and thereby focusses on feature engineering, data quality, and visualization. To facilitate the deployment of prediction algorithms in situations without complete cases, we propose an adaptation to the k-nearest neighbours imputation algorithm which uses expert knowledge. Overall, our proposed method correlates strongly with eventual rider performance and can aid scouts in targeting young talents. On top of that, we introduce several model interpretation tools to give insight into which current starting professional riders are expected to perform well and why.

State transitions among groups of cyclists in cycling points races

This study analyzed the global state of groups formed by cyclists competing in three different points races and considered the behaviour of individual cyclists in those races. We measured the time difference between the front cyclist and the other cyclists when they crossed the centrelines of home and back straights in order to quantify the global configuration of cyclists in terms of their density and features of states, extracted using principal components analysis (PCA). We examined whether the group separation and group density that characterize the cycling race can be extracted by PCA. We interpreted the PCA results to explain the separation and density of the group using the first and second principal components. Then, we defined the state of configuration of the cyclists in each lap in the plane of the first and second principal components. Subsequently, the state transition probabilities were obtained. States 1, 2, 3, and 4 corresponded to the third, second, first, and fourth quadrants, respectively. State 1 represented a state comprising one dense group, state 2 represented one stretched group, state 3 represented a divided group, and state 4 represented an escape group far from a single dense group.Highlights An approach to understand the collective behaviour of cycling points races through principal component analysis was effective for quantifying the configuration of the cyclists.Principal component analysis of the global configuration of the cyclists in the points races revealed the fission-fusion dynamics was characterized by two components. The density of a group and number of groups, and transitions among four states was defined by these two components.State transition probabilities indicate that the group separation states were more frequent in the latter half of the sprint interval, and it was difficult to re-combine the separated groups into one.The riders and coaches need to be aware of the stretching and separation of the group, even if it does not occur immediately before the sprint as the positioning of a cyclist in the group would be important at that time.