Footwear Science struggles at identifying the unknown effects of footwear interventions on the biomechanics of running. Those struggles may partly be explained by methodological and statistical issues, such as using discrete parameters to reduce the dimensionality of time series data or testing against unspecific null hypotheses. This paper evaluates a method to address some of these deficits by applying a time series-based machine learning approach (k-nearest neighbour classification (kNN), Euclidean distance). As an example, we used a data set containing a largely unknown footwear effect: The relationship between subjective comfort and the rollover movement during running. For this purpose, we collected kinematic data and subjective ratings from 22 runners running two times each in both their most comfortable and uncomfortable shoe selected from a pool of different neutral running shoes. The kNN algorithm performed highly accurately in distinguishing between the kinematics of different shoes. We also found that larger differences in subjective comfort perception between two shoes were associated with higher classification rates. The kNN method, however, does not allow for conclusions to be drawn about the causality or quality of the observed changes within the kinematic trajectories. Accordingly, we see the main scope of the presented method primarily in exploring complex data sets in cases where no meaningful hypothesis about the presumed biomechanical effects can be derived a priori. The subsequently acquired knowledge may be used to design larger-scale studies in which these findings can be investigated more systematically by means of more specific hypotheses and better estimates of sample sizes, respectively.
© Copyright 2020 Footwear Science. Taylor & Francis. Kaikki oikeudet pidätetään. / Kääntänyt https://www.DeepL.com/Translator