VISTA-SSM: Varying and irregular sampling time-series analysis via state-space models.

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• Healthcare and psychology data often feature irregular sampling.
• Standard time series clustering methods struggle with varying sampling rates.
• Unsupervised group identification is crucial for complex datasets.

Purpose of the Study:

• Introduce Varying and Irregular Sampling Time-series Analysis (VISTA).
• Develop a clustering approach for multivariate, irregularly sampled time series.
• Provide a flexible framework for diverse time series dynamics.

Main Methods:

• Utilize a parametric state-space mixture model.
• Adapt linear Gaussian state-space models (LGSSMs).
• Employ an expectation-maximization algorithm for parameter fitting.

Main Results:

• VISTA effectively handles multivariate time series with irregular sampling.
• The method demonstrates accuracy on simulated and real-world data.
• VISTA outperforms comparable standard time series clustering methods.

Conclusions:

• VISTA offers a robust solution for clustering irregularly sampled time series.
• The approach is versatile across various domains like healthcare and sensor data.
• An open-source Python implementation is available.