Exploring drivers and policy enablers of citizen engagement in renewable energy projects through explainable AI

Community Renewable Energy (CRE) projects play a critical role in the global clean energy transition by enabling citizens to collectively own and manage energy systems. Despite growing interest, the factors driving participation—and the extent of their influence—remain underexplored. Most prior studies conceptualize participation narrowly, overlook the relative importance of drivers, and rely on traditional statistical methods that struggle to capture complex behavioral patterns. Addressing these gaps requires approaches suited to high-dimensional data with nonlinear interactions. Machine Learning (ML) offers the flexibility to model such complexity, while explainable Artificial Intelligence (XAI) techniques provide transparency by interpreting predictions independently of the underlying algorithm—essential for generating policy-relevant insights. Using Australia as a case study, this research addresses three questions: whether ML models can accurately predict citizens’ multi-role participation levels; how robust model-agnostic explanations are when applied to heterogeneous behavioral data; and what factors drive varying levels of engagement. We surveyed 875 residents on their participation across ten CRE roles—from investing to volunteering and advocacy—synthesizing responses into a composite involvement score categorized as low, medium, or high. Classification algorithms including eXtreme Gradient Boosting (XGBoost), Multi-layer Perceptron (MLP), and Keras-based Sequential models were paired with SHapley Additive exPlanations (SHAP)—a model-agnostic XAI method—using Tree SHAP, Kernel SHAP, and Deep SHAP variants for interpretation. To evaluate explanation robustness, we developed two novel metrics: the Stability Correlation Index (SCI) and the Explanation Integrity Metric (EIM), which introduce targeted perturbations to key features. XGBoost with Tree SHAP achieved the highest stability, with SCI above 99% and EIM near zero at moderate perturbation levels. Key findings reveal that plans to adopt renewable technologies polarize participation into high or low levels, while perceptions of corporate-owned projects act as a double-edged factor driving both engagement extremes. Increased awareness of policy risks motivates moderate participation, suggesting opportunities to leverage risk communication. These insights offer practical strategies for policymakers and project developers seeking to broaden and deepen citizen engagement in CRE initiatives. • Developed SCI and EIM metrics to measure explainers’ sensitivity. • Employed XGBoost and Tree SHAP to provide robust explanations. • Plans to adopt renewable technologies polarize participation into high or low levels. • Belief in corporate-owned projects drives both high and low engagement extremes. • Policy risk awareness motivates moderate participation, challenging deterrence views.