The purpose of this study was to develop a method which uses positioning data to analyse player-specific skating characteristics and to investigate the possibility to use machine learning to generate new individual and game-specific training drills. A real-time local positioning system was used to collect positioning data from a professional ice hockey team in the Swedish Hockey League. Positioning data and video were synchronized, and nine different skating characteristics were manually identified and tagged for two forwards. A cost function was developed to generate individual, continuous skating sequences and to create new individual and game-specific skating drills. Skating forward was the most commonly used skating characteristic for both player but the numbers of times the players used the different skating characteristics varied, (H(8)=23.2, p=0.003). The number of skating characteristics between the two players varied, (χ2 (8, N=688) =3 4.0, p<0.001) as well as the time spent within each skating characteristic (p<0.001). The presented method can be used for performance analysis and shows promising results for creating individual and game-specific tests and training drills for ice hockey players, based on individual and game-specific skating characteristics