Detecting Multiple Step Changes Using Adaptive Regression Splines with Application to Neural Recordings

Time series produced by dynamical systems as frequently the case in neuroscience are rarely stationary but often exhibit quite abrupt changes due to bifurcations or other dynamical phenomena. A plethora of methods for detecting such changes in time series statistics, commonly called change point analysis, have been developed over the years, in addition to test criteria to evaluate change point significance. Issues to consider when developing such methods include computational demands, difficulties arising from either limited amount of data or a large number of covariates, and arriving at statistical tests with sufficient power to detect as many change points as contained in potentially high-dimensional data sets. Here, a general method called Paired Adaptive Regressors for Cumulative Sum (PARCS) is developed for detecting multiple change points in multivariate time series. The method's flexibility to incorporate useful features from other change point detection techniques is highlighted. The advantages of PARCS over existing approaches are demonstrated through a series of simulation experiments, followed by a real data application to neural recordings from rat medial prefrontal cortex during learning.