Finding the best fit of rowers for a crew boat is a challenging task. Each rower has a unique technique and the ability to adapt this to a crew varies from person to person. Currently, subjective evaluations and qualitative measures are the main methods used to try to put the fastest crew together. To make the process more accurate and objective we introduce 177 performance metrics to quantify some of the measureable aspects of rowing technique. We then present two data-driven approaches to select the most relevant features that 1) make individual rower's technique unique and 2) correlate most strongly to boat speed. The first approach uses sequential forward feature selection to identify the features that are most discriminative for individual rowers in crew boats. These features make the unique biomechanical fingerprint of each rower. We recorded a dataset with four world-class female rowers racing in double sculls in different crew combinations. We identified the "Finish Slip" as the most discriminative feature. A rower identification classifier based solely on this feature scored an accuracy of 74.6%. Applying one or two additional features this accuracy improved to 90.7% or 95.6% respectively. In a second approach we proposed linear regression analysis to identify the features that most strongly correlate to boat speed. For the given dataset, a subset of five performance metrics proved sufficient to build a linear model that predicts the boat speed with a root mean square error of less than 0.087 m/s.
© Copyright 2015 International Journal of Computer Science in Sport. Sciendo. Kaikki oikeudet pidätetään. / Kääntänyt https://www.DeepL.com/Translator