Identifying variable importance on mixed collinear building energy factor space by stochastic multi-statistic sensitivity analysis on TOPSIS with random decision forest

Accurately quantifying variable importance (VI) is essential to the success of building energy efficiency for mitigating climate change. Extant methods fail to generate reliable and interpretable VI outcomes on mixed collinear energy factor space. This paper develops an alternate stochastic sensitivity analysis method to fully integrate categorical and interactional effects while handling collinearity for VI in buildings. While data-based sampling is utilized for incorporating global factor interactions through differential analysis, it uses differentiative levelling to explicitly characterize full energy effects of building factors. To cope with potential collinearity, the ensemble random decision forest is employed to acquire stochastic energy responses by bootstrapping and randomized sub-spacing. On a multi-statistic metric gauging input-output sensitivity from diverse angles, the rank-oriented compensatory Technique of Order Preference Similarity to the Ideal Solution (TOPSIS) is adopted to quantify VI with objective weighting. Case results regarding 191 U S. houses validated the efficacy of presented approach with evidenced robustness. This approach can aid in building energy efficiency by supplying precise VI information on complete factor spaces while retaining transparency.