Structural equation modeling, machine learning, neglected tropical diseases

Machine learning-enhanced SEM for spatial and time-varying causal modeling of NTDs

Neglected Tropical Diseases (NTDs) affect over one billion people in low-resource settings, with their epidemiological dynamics driven by a complex interplay of social, environmental, and behavioral factors that vary across time and space. Structural Equation Modeling (SEM) offers a robust and interpretable framework for capturing the causal relationships between observed and latent variables. However, classical SEM approaches rely on linearity and stationarity assumptions, which limit their capacity to model the nonlinear and dynamic nature of real-world epidemiological processes.
This work proposes an innovative extension of the SEM framework by integrating machine learning (ML) techniques, such as neural networks and random forests, to enhance causal model flexibility and predictive accuracy. The approach strategically combines SEM for latent score estimation and causal structure identification, with ML algorithms to model residual patterns, nonlinear interactions, and spatio-temporal dynamics.
Expected outcomes include significantly improved prediction of NTD prevalence, better capture of latent nonlinearities and geographic/temporal variability, and empirical validation of a hybrid SEM-ML framework capable of generalizing to diverse public health contexts. This integrated approach offers promising prospects for dynamic disease modeling, epidemiological surveillance, and data-driven health intervention planning—bridging classical causal inference with modern machine learning techniques.