Testing and mapping non-stationarity in animal behavioral processes: a case study on an individual female bean weevil.

ABSTRACT

Statistical analysis based on two characteristics of a small-world network, and on Lempel-Ziv's measure of Kolmogorov-Chaitin's algorithmic complexity are first proposed to scan through an individual behavioral sequence for possible existence of non-stationarity. Due to fixed window width, these tests have drawbacks in mapping out regions of non-stationarity. A non-parametric approach based on sparse coding schemes is employed to segment the whole behavioral sequence into unequal length segments, thus resultant avoiding further efforts for grouping. Then attempts are made to entangle the resultant segmentation with other non-local behavioral patterns onto such sequence to ascertain that the non-stationarity corresponds to a sequence of different categories of underlying driving force. It is of potential importance that this segmentation, represented by a hierarchy of code sequences, provides a natural platform for detecting intrinsically coherent behavioral patterns based on continuously recorded data. Illustrations throughout the developments are made exclusively on data encoded from a nearly 4-h video-recording of a female bean weevil's behavior.