Segmentation of partial least squares structural equation modelling using kernel K-means clustering (PLS SEM KKC).

Area of Science:

• Statistical Modeling
• Machine Learning Applications
• Multivariate Data Analysis

Background:

• Partial Least Squares Structural Equation Modeling (PLS SEM) is widely used but limited by unobserved heterogeneity.
• Existing PLS SEM segmentation methods rely on linear clustering, failing to capture non-linear residual patterns.
• Unobserved heterogeneity can lead to inaccurate and unreliable PLS SEM models.

Purpose of the Study:

• To propose and evaluate a novel non-linear segmentation method for PLS SEM, named PLS SEM Kernel K-Means Clustering (PLS SEM KKC).
• To address the limitation of unobserved heterogeneity in PLS SEM by incorporating kernel-based clustering.
• To enhance the accuracy and reliability of PLS SEM models through effective segmentation.

Main Methods:

• Developed PLS SEM Kernel K-Means Clustering (PLS SEM KKC) by integrating kernel-based clustering with PLS SEM.
• Segmentation was performed based on the non-linear residual values from measurement and structural models of a global PLS SEM.
• Employed clustering to group observations with similar residual patterns into homogeneous segments.

Main Results:

• The PLS SEM KKC method significantly improved model accuracy compared to the global model.
• R² values increased from 51.1% (global model) to 93.9% (k=2) and 97.5% (k=3) in segmented clusters.
• The substantial increase in local R² demonstrates the successful overcoming of unobserved heterogeneity.

Conclusions:

• PLS SEM KKC is a recommended new method for PLS SEM segmentation.
• The method effectively captures non-linear residual patterns, successfully addressing unobserved heterogeneity.
• PLS SEM KKC leads to more accurate and robust PLS SEM models by creating homogeneous segments.