Methods / Statistics
Functional principal component analysis (FPCA) can be used to extract key features from time series data for use in statistical models. This study evaluated time normalisation in combination with curve registration prior to performing FPCA. Using vertical ground reaction force data from countermovement jumps, evaluation was based on linear regression for predicting peak power and jump height, and logistic regression for classifying jump type (arm swing or not). Datasets not subject to time normalisation generally produced better results with the highest accuracy being achieved when using registration with peak power as a landmark (peak power R2 = 99.3%, jump height R2 = 94.9%). Classification of jump type benefited in some cases from registration (87.0% to 91.2%). These techniques could be applied to data from wearable sensors to improve prediction and classification.
White, Mark; Bezodis, Neil; Neville, Jono; and Summers, Huw
"FORCE-TIME CURVE ALIGNMENT FOR FUNCTIONAL PRINCIPAL COMPONENT ANALYSIS IN VERTICAL JUMPING,"
ISBS Proceedings Archive: Vol. 38
, Article 82.
Available at: https://commons.nmu.edu/isbs/vol38/iss1/82