Singular Spectrum Analysis as a data-driven approach to the analysis of motor adaptation time series

[Display omitted] •The utility of SSA to decompose and assess motor adaptation time series is tested.•Trend reconstruction performance of SSA was assessed and compared with filter and fitting methods.•SSA reconstructed the simulated global trends more accurately than the filter and fitting methods.•SSA was able to reconstruct global trends from idiosyncratic split-belt adaptation time series. Motor adaptation is a form of learning to re-establish desired movements in novel situations. To probe motor adaptation, one can replicate such conditions experimentally by imposing a sustained perturbation during movement. Exposure to such perturbations initially causes an abrupt change in relevant performance variables, followed by a gradual return to baseline behaviour. The resulting time series exhibit persistent properties related to structural changes in underlying motor control and transitory properties related to trial-to-trial variations. The global trend, signifying the structural change, is often assessed by averaging the time series in predefined bins or nonlinear model fitting. However, these methods to study motor adaptation require a priori decisions to produce accurate fits. Here, we test a data-driven approach called Singular Spectrum Analysis (SSA) to assess the global trend. In SSA, we first decompose the adaptation time series into components that represent either a global trend or additional variations, and secondly, select the component(s) corresponding to the global trend using spectral analysis. In this paper, we will use simulated data to compare the reconstruction performance of SSA with often applied filter and fitting methods in motor adaptation studies and apply SSA to real data obtained during split-belt adaptation. In the simulations, we show that SSA reconstructed the fast-initial component and entire global trends more accurately than the filtering and fitting methods. In addition, we show that SSA also successfully reconstructed global trends from real data. Therefore, the SSA might be useful in motor learning studies to decompose and assess adaptation time series.