Figure 7
From: Quantification of early learning and movement sub-structure predictive of motor performance
Temporal profile and support vector machine analysis of intra-session features. (A) Temporal profile of variance of vertical position of paw calculated for 1 s time intervals (blue line). The red line arises from the treatment of the unmodified data with a function that assigns a value of zero to values that deviate 6 or more standard deviations from the mean. A 3rd degree polynomial least-squares best-fit line (red) illustrates trends across seconds. The last 5 s have been omitted to avoid including values corresponding to a falling mouse. (B) Best fit trend lines for several parameters illustrate a change in the last 20 s before the mouse falls off the rotarod. The last 5 s have been omitted to avoid including mouse falls. A = Lowermost paw position, B = median paw position, C = Downward paw speed (negative values indicate downward movement), D = Downward paw acceleration (negative values indicate downward movement), E = Upward paw speed (negative values indicate downward movement), F = Upward paw acceleration (negative values indicate downward movement), G = Variance of the vertical position of the paw, H = Approximate entropy of the vertical position of the paw. (C) Support Vector Machine (SVM) classifier. Left panel: Two-dimensional (2D) example scatter plot of vertical paw position and approximate entropy calculated for the first 20 one-second time intervals contrasted with 20 one-second time intervals at the end of the rotarod session. Each dot corresponds a 1-s time interval or bin. The last 5 s have been omitted as above. The data-driven decision boundary distinctly separates the two classes of mice. Right panel: Receiver operating characteristic (ROC) curve for classifying the first and last 20-s time-bin data in separate clusters for the mouse depicted in the left panel. The classifier analysis was conducted in 6-dimensional space on vertical and horizontal paw speed, acceleration, and variance of paw position calculated for 21-s time bins. (D) Scatter plots for different mice illustrating separation of features corresponding to the first and last 20 s of a rotarod session. The misclassification rate (MR) calculated from an SVM classifier is indicated for each graph.
