Landmark registering waveform data improves the ability to predict performance measures

The purpose of this study was to investigate the benefit of landmark registration when applied to waveform data. We compared the ability of data reduced from time-normalised and landmark registered vertical ground reaction force (vGRF) waveforms captured during maximal countermovement jumps (CMJ) of 53 active male subjects to predict jump height. vGRF waveforms were landmark registered using different landmarks resulting in four registration conditions: (i) end of the eccentric phase, (ii) adding maximum centre of mass (CoM) power, (iii) adding minimum CoM power, (iv) adding minimum vGRF. In addition to the four registration conditions, the non-registered vGRF and concentric phase only were time-normalised and used in subsequent analysis. Analysis of characterising phases was performed to reduce the vGRF data to features that captured the behaviour of each waveform. These features were extracted from each condition's vGRF waveform, time-domain (time taken to complete the movement), and warping functions (generated from landmark registration). The identified features were used as predictor features to fit a step-wise multilinear regression to jump height. Features generated from the best performing registration condition were able to predict jump height to a similar extent as the concentric phase (86-87%), while all registration conditions could explain jump height to a greater extent than time-normalisation alone (65%). This suggests waveform variability was reduced as phases of the CMJ were aligned. However, findings suggest that over-registration can occur when applying landmark registration. Overall, landmark registration can improve prediction power to performance measures as waveform data can be reduced to more appropriate performance related features.