Clustering methods for categorical time series and sequences : a scoping review

Area of Science:

• Data Science
• Bioinformatics
• Epidemiology
• Sociology
• Marketing Analytics

Background:

• Categorical time series (CTS) data are prevalent across diverse scientific disciplines, including epidemiology, sociology, biology, and marketing.
• Clustering these complex data structures presents unique challenges, such as variable sequence lengths, multivariate observations, and missing data.

Purpose of the Study:

• To systematically review and categorize existing clustering methods specifically designed for categorical time series (CTS) data.
• To provide a framework for selecting appropriate CTS clustering methods based on specific data characteristics and research objectives.

Main Methods:

• A comprehensive literature search was conducted across PubMed, Web of Science, and Google Scholar up to November 2024.
• Identified CTS clustering methods were classified into three main families: distance-based, feature-based, and model-based.
• Methods were evaluated based on their ability to handle common challenges like variable sequence length, multivariate data, and missing values.

Main Results:

• Out of 14,607 retrieved records, 124 articles describing 129 distinct CTS clustering methods were included in the review.
• Distance-based methods, particularly those employing Optimal Matching, were the most frequently proposed (56 methods).
• Model-based methods (28) demonstrated greater capacity for complex data structures, while feature-based methods (45) offered better scalability at the expense of flexibility.
• Public implementations were available for fewer than half of the reviewed methods.
• A searchable web application was developed to facilitate method selection.

Conclusions:

• Categorical time series clustering methods exhibit significant heterogeneity in their underlying assumptions, capabilities, and scalability.
• While distance-based methods are prevalent, model-based approaches offer more sophisticated modeling potential for complex data.
• The developed typology and web application serve as valuable resources for researchers navigating the diverse landscape of CTS clustering techniques.