A MULTIMODAL STRUCTURAL EQUATION MODELLING INTEGRATION FOR ASSESSING PHYSIOLOGICAL AND PSYCHOLOGICAL OUTCOMES IN AUTISM SPECTRUM DISORDER

Abstract

This paper presents a multimodal computational framework that integrates electroencephalogram (EEG) signals with Bio-Well gas discharge visualization (GDV) for assessing neurophysiological regulation during structured auditory stimulation. Traditional approaches to physiological state classification often sacrifice interpretability for accuracy. A hybrid architecture combining Partial Least Squares Structural Equation Modelling (PLS-SEM) with machine learning classification is proposed. Objective 1 applied PLS-SEM to Bio-Well Gas Discharge Visualisation (GDV) measurements from 30 individuals, modelling intervention-induced changes through three latent constructs: Physiological Regulation (PR), Energy-Emotion Coherence (EEC), and Psychological Well-Being (PWB). Objective 2 validated the structural model on an independent EEG dataset containing 90 subjects and trained an XGBoost classifier for ASD versus control discrimination. Structural analysis confirmed a strong PR-to-EEC association (β = 0.990, p < 0.001) and an EEC-to-PWB association (β = 1.078, p < 0.001), while the direct PR-to-PWB path was non-significant (β = -0.082, p = 0.703), establishing full mediation through EEC. Predictive relevance was confirmed via blindfolding with Q² values of 0.743 for EEC and 0.812 for PWB. The XGBoost classifier achieved 92.2% accuracy (AUC-ROC = 0.97) on the EEG dataset with a ten-fold cross-validation accuracy of 91.8 ± 2.3%. These results offer a data-driven, interpretable foundation for personalised non-pharmacological intervention assessment.