Hussein El Amouri
← Back to portfolio

Research project · Published research

Constrained DTW Preserving Shapelets

Learning time-series representations that incorporate expert constraints while preserving Dynamic Time Warping relationships.

Time seriesConstrained clustering
Time-series data and expert pairwise constraints inform a DTW-preserving shapelet representation, which is used for clustering.
Conceptual illustration. Time-series data and expert pairwise constraints inform a DTW-preserving shapelet representation, which is used for clustering. Open full size ↗

CDPS integrates must-link and cannot-link constraints into a shapelet-based time-series representation. It supports clustering with expert knowledge and transforming out-of-sample data.

See the Pattern Recognition paper for the method and evaluation.

Constrained DTW Preserving Shapelets