Temporally aligned segmentation and clustering (TASC) framework for behavior time series analysis.

Area of Science:

• Neuroscience
• Behavioral Science
• Data Science

Background:

• Behavioral data possesses intricate spatiotemporal structures, often comprising discrete sub-behaviors or motifs.
• Accurate segmentation and clustering of continuous behavioral data are crucial for motif discovery.
• Current automated behavior quantification tools often lack the precision and scalability needed for experimental and clinical research.

Purpose of the Study:

• To develop a generalized framework for precise behavior motif segmentation and clustering.
• To improve the analysis of complex, high-dimensional time-series behavioral data.
• To provide a robust foundation for advanced behavior research and related fields.

Main Methods:

• Proposed a generalized framework employing an iterative approach to refine segmentation and clustering.
• Introduced Temporally Aligned Segmentation and Clustering (TASC), utilizing temporal linear alignment for motif distance computation and alignment.
• Implemented an alternating-step process involving neighbor evaluation against cluster centroids and re-clustering of selected segments.

Main Results:

• Demonstrated enhanced segmentation and clustering performance on semi-synthetic and real-world datasets.
• Validated the framework's effectiveness in experimental and clinical settings.
• Showcased TASC's capability for precise identification and analysis of recurring behavior motifs.

Conclusions:

• The TASC framework offers a significant advancement in analyzing complex behavioral time-series data.
• It provides a more accurate foundation for subsequent research in behavior quantification.
• The methodology is adaptable for extending existing tools and applicable to other domains requiring high-precision time-series segmentation.