Identifying ski roughness using data driven approaches

In skiing sport, snow friction is a crucial factor in determining the ski roughness that can produce high speed and quick finishing time for a skier. However, snow friction is influenced by many factors associated with weather and snow conditions which affect the choice of the optimal ski’s roughness. This paper proposes an ensemble learning system that can accurately recommend the best ski roughness under different weather conditions. The data used in this study is a unique data set that has been collected from field tests and competitions. Though this data set recorded information about ski treatment and weather conditions over a 10-years period, it is affected by noise and outliers, and it has an imbalanced distribution in the ski roughness classes. This work addresses these challenges in the data by applying preprocessing techniques and class balancing strategies. Furthermore, correlation and clustering approaches are employed to identify redundancies in the data and to recognise the subsets of weather conditions that have the highest influence on the selection of the ski roughness. Using the resultant clusters, an ensemble system is introduced to recommend the most suitable skis roughness for a given weather condition. This system can be used as a guiding tool in skiing competitions to aid technicians in choosing the skis roughness. The results showed that air and snow temperatures as well as snow humidity have the highest impact on the choice of the ski roughness.