Enhancing retrofitting measure assessment in residential buildings through a Shapley Additive Explanations interaction weighting method

The comprehensive assessment of retrofitting techniques is essential for improving energy efficiency and sustainability in residential buildings. This study presents a novel hybrid decision-making framework that integrates explainable artificial intelligence (XAI) with advanced multi-criteria evaluation techniques. The method uses the Shapley Additive Explanations (SHAP) interaction matrix to obtain weights for each criterion. The matrix reports both the direct effect of each variable and the interaction terms produced by the model. These values tie the weighting scheme to the behaviour recorded in the dataset. The evaluation based on relative utility and nonlinear standardisation (ERUNS) method ranks the retrofit measures. It places all criteria on a shared numeric scale through non-linear standardisation and applies a utility rule that produces one score for each option. The use of SHAP interaction weights together with the ERUNS scoring rule produces a ranking system tied to recorded data and stated weight choices. The proposed methodology shows how each retrofit option responds to the interaction structure of the criteria and the utility rule used to score them. The case study on residential buildings shows how this method works on an actual dataset and produces a defined order of retrofit measures. The results show which options rise or fall under the given criteria and weight settings.