A Gentle Introduction and Application of Feature-Based Clustering with Psychological Time Series.

Area of Science:

• Psychology
• Data Science
• Computational Social Science

Background:

• Psychological research increasingly involves complex time series data to understand participant development.
• Existing clustering methods often fail to incorporate meaningful dynamic measures (e.g., instability, inertia) due to model assumptions.
• There is a need for flexible and transparent clustering approaches tailored for psychological time series data.

Purpose of the Study:

• To propose and illustrate feature-based time series clustering for psychological research.
• To enable clustering of participants based on meaningful dynamic developmental measures.
• To address challenges in ecological momentary assessment (EMA) data, including multivariate conceptualizations, missingness, and non-stationarity.

Main Methods:

• Feature-based time series clustering is proposed, utilizing common clustering algorithms on extracted dynamic measures.
• Key steps include input selection, feature extraction, feature reduction, clustering, and cluster evaluation.
• The approach is demonstrated using real-world EMA data.

Main Results:

• Feature-based clustering successfully clusters participants based on dynamic developmental measures.
• The method effectively handles complexities common in psychological time series data, such as missing values and changing trends.
• Practical guidance and code are provided for implementation.

Conclusions:

• Feature-based time series clustering offers a flexible, transparent, and effective method for analyzing psychological developmental data.
• This approach overcomes limitations of traditional methods, allowing for richer insights into individual differences.
• The proposed methodology and accompanying resources facilitate advanced analysis of complex psychological time series.