HeteroX: An interpretable Dual-tier Heterogeneous Graph for sustainable investment decisions

Understanding sustainable investment behavior is crucial to advancing low-carbon transitions, but modeling it remains challenging due to the emotionally driven, context-dependent, and socially embedded nature of individual decision-making. This challenge is particularly evident in Community Renewable Energy (CRE) initiatives, which rely heavily on citizen investment yet continue to face persistently low participation rates. Classical predictive models—typically based on flat, decontextualized feature sets—fail to capture the structural interdependencies that shape such decisions, including social identity, spatial belonging, and institutional trust. In this study, we introduce a graph-based framework that encodes these interdependencies through a Dual-tier Heterogeneous Graph (DHG), which integrates individuals and structured survey features via intra- and inter-layer links. A Graph Neural Network (GNN) trained on this architecture predicts investment willingness. To make these predictions intelligible, we develop HeteroX, a novel post hoc interpretability method tailored to heterogeneous graphs. Applied to a real-world survey of 875 Australian citizens on CRE investment, the DHG–HeteroX pipeline achieved 94% accuracy, compared to 50%–80% accuracy for baseline classifiers. For interpretability, HeteroX reached fidelity of 0.92, completeness of 0.96, and consistency of 0.91, outperforming state-of-the-art explainers such as Integrated Gradients, GraphShap, and GNNExplainer. The framework further highlighted affective resonance, economic empowerment, and community connectedness as the strongest predictors of investment willingness. These findings suggest that interpretability in relational models is itself relational, and that structure-aware explanations can illuminate the logic underpinning socially embedded decisions. • Investment willingness within Community Renewable Energy initiatives modeled. • A dual-tier heterogeneous graph (DHG) captures social, spatial, institutional drivers. • Introduces HeteroX: the first post hoc explainer for DHG architectures. • Outperforms state-of-the-art in fidelity, sparsity, and interpretability. • Reveals emotional and social embedding as key drivers of green investment.