The main goal of this thesis is to expand data analytics to the sport of rowing, using biomechanical data collected on rowers' power profiles and connectivity to the water. Using a dataset collected on 115 race pieces done between heavyweight rowers in the eight category, we aim to relate kinetics (power application) to performance, in a more mathematical and rigorous way to quantify what should be the main focus of athletes going forward. Regression Models, Trees, and Neural Networks were applied to predict race outcomes and quantify feature contribution and specific aspects of the stroke such as length, angles, and power curves. In doing so we found Extreme Gradient Boosted Trees most suitable for race outcome predictions (Accuracy = 69%), and features (SwivelPower, Work PC Q1, and Curve Smoothness) that represent power and curvature most dominant in their contribution to prediction models. We have further analyzed force curve variables to understand the difference in shape and smoothness between winning and losing crews and found that the shape of these curves differs most around the perpendicular.
© Copyright 2023 Veröffentlicht von Princeton University. Alle Rechte vorbehalten.