THE DETERMINATION OF BREAK POINTS IN TIME SERIES DATA

Abstract

The aim of this paper was to demonstrate three possible methods for the quantification of breakpoints. Data were the elbow angle during the pull phase of 10 strokes of ergometer rowing. First, a piecewise linear regression of two segments separated by a breakpoint was applied iteratively whereby the minimum pooled squared error of both regressions was the breakpoint. Second, the last local maxima, identified through zero-crossings of the first derivative, was the breakpoint. Thirdly, the last time point at which the rate of change of the joint angle data crossed a threshold was identified as the breakpoint. Each of these methods was checked against visual determination of the breakpoints. The threshold method was closer to the breakpoint as identified by visual inspection than were the linear regression and local maxima methods, and appeared suitable for this application.