Short-Term Scientific Mission

Optimization Algorithm for Automatic Detection of Structural Relations in Structural Equation Modeling

Main theme: Algorithm Selection and Configuration
Grantee: Gloria Pietropolli, Universita degli Studi di Trieste, Trieste, Italy
Host: Laura Trinchera, NEOMA Business School, Paris, France

Start date: 2024-09-21
End date: 2024-09-28
Awarded: 2024-09-03
Report approved: 2024-10-28

Description

Structural Equation Modeling (SEM) is widely used in social sciences, psychology, and education to validate theoretical models by analyzing relationships between observed and latent variables. Traditional SEM approaches rely on manually specified models, but recent advances underscore the need for automation to enhance efficiency and accuracy. This STSM focuses on developing an optimization algorithm, specifically a Genetic Algorithm (GA), to automate SEM model selection. The goal is to optimize model parameters based on the Akaike Information Criterion (AIC), providing a systematic, data-driven approach to model configuration and addressing the gap in automated methodologies for composite-based SEMs.

Achievements

This STSM explored the use of Genetic Algorithms for SEM model selection. The focus was on developing an efficient search strategy to identify well-fitting SEM configurations while reducing the need for manual specification. During the visit, key aspects of GA-based optimization for SEMs were examined, including parameter tuning and feasibility assessment. An initial prototype was developed and tested, providing insights into the strengths and limitations of this approach. Discussions helped refine the problem formulation and align optimization objectives with practical SEM challenges, highlighting directions for further improvement.

Beyond the technical work, the STSM established a collaboration between the University of Trieste and NEOMA Business School, creating a foundation for continued research. A paper detailing the methodology and initial results is in preparation, and future efforts will focus on refining the optimization strategy and expanding experimental validation.