Temporally Aligned Segmentation and Clustering (TASC) of Behavioral Time Series
Ekaterina Zinkovskaia, Orel Tahary, Izhar Bar‐Gad · 2023
Behavior consists of a series of repeating yet variable discrete motifs across various timescales. We introduce a framework for temporally aligned segmentation and clustering (TASC) of behavioral time series. TASC is designed to extract such motif recurrences in high temporal resolution. This framework operates iteratively in two steps: (1) embedding of time series segments and calculation of linearly aligned distances within the clustered space, and (2) recalculating of the clustered space after alignment. We evaluated TASC on a semi-synthetic experimental and a clinical dataset, and it demonstrated enhanced segmentation performance. TASC may be applied to other domains where analysis of recurring time series patterns with high temporal precision is needed.Clinical Relevance: This framework enables identifying the temporal structure of discrete pathological behaviors, such as tics properties in Tourette’s syndrome.