A method for locating gradual changes in time series.

A variation to the change-point problem is addressed. The classical problem involves locating abrupt changes in the mean value of a signal. In contrast, a generalisation to gradual changes with constant speed is considered, which frequently occurs in biomedical signal-processing tasks. Formulas are derived that are easy to adopt by application scientists. The estimation quality is investigated theoretically and using Monte Carlo simulations. Results show that the change-point estimates were close to their true values. In contrast, a systematic error of approximately half the change duration occurred when the gradual nature of the signal was neglected.