CSMVC: A Multiview Method for Multivariate Time-Series Clustering.

Area of Science:

• Data Mining
• Machine Learning
• Time Series Analysis

Background:

• Multivariate time-series (MTS) clustering is crucial for data analysis.
• Existing methods face challenges like high computational cost and information loss.
• Most current approaches are single-view, neglecting the advantages of multiple data perspectives.

Purpose of the Study:

• To develop an effective multiview clustering method for MTS data.
• To address limitations of existing single-view clustering techniques.
• To propose a novel approach that leverages both shared and unique information across multiple views.

Main Methods:

• A consistent and specific non-negative matrix factorization-based multiview clustering (CSMVC) method is proposed.
• A multilayer graph represents MTS data, generating multiple views via subspace techniques.
• A novel non-negative matrix factorization (NMF) approach explores view-consistent and view-specific information simultaneously.

Main Results:

• The CSMVC method demonstrated superior performance across 13 benchmark datasets.
• The approach effectively handles the unique data structure of MTS.
• Experimental results show significant improvements over state-of-the-art algorithms.

Conclusions:

• The proposed CSMVC method offers a robust solution for MTS clustering.
• It effectively integrates information from multiple views for enhanced accuracy.
• CSMVC represents a significant advancement in multiview clustering for time-series data.