Sustainable Cities and Building Energy Efficiency
This project develops an explainable machine learning framework for analyzing and reducing building heating loads. The workflow combines automatic model selection with SHAP-based explainability to identify the key building characteristics influencing energy demand. Beyond prediction and interpretation, the framework generates actionable counterfactual recommendations using DiCE and NSGA-II under feasibility constraints. The two approaches are evaluated in terms of validity, sparsity, proximity, and diversity, enabling a comparison between conventional counterfactual explanations and multi-objective optimization for practical building energy-efficiency interventions.
More work
