Abstract
In recent times, smartphones and watches have facilitated researchers in collecting real-time through ecological momentary assessment (EMA). This data enables the exploration of human behavior, emotions, and psychological processes in their natural environments. Clustering is one of the techniques to identify different subgroups of individuals based on extracted information from EMA items. This study focuses on model-based clustering, which involves extracting information/coefficients from linear or non-linear machine-learning models and inputting extracted coefficients to the traditional clustering algorithm, such as K-Means or hierarchical clustering. Primarily, linear models are used that utilized linear parameter estimates as coefficients and non-linear models, such as decision models, where feature importance is used instead of coefficients. Results demonstrate the predictive performance of linear and non-linear models as well as the clustering performance and analysis of distinct clusters of individuals. Moreover, Linear Lasso Regression (Lasso) with K-Means illustrates a balance cluster distributions for 2 clusters, highlighting special characteristics of individuals within each cluster. Nevertheless, Vector Autoregression (VAR) with K-Means exclusively assigns cluster based on a single attribute of individuals. Generally, Model-based clustering techniques, such as VAR and Lasso K-Means demonstrate the ability to capture behavioral patterns among individuals, which are represented as coefficient matrices, for grouping them into distinct clusters.
Figures
