Integrating STIRPAT - Adaboost and data-driven for influencing factors analysis and scenario prediction of carbon emission during building materialization stage

The issue of carbon emissions (CE) from buildings has come into focus. Among them, CE during the building materialization stage (BMS) account for a large proportion. This paper takes the past 20 years of China's carbon emission data during the BMS as a case study. Firstly, we select influencing factors based on multi-source data according to IPAT identification, then use a machine learning method to conduct a secondary screening of the influencing factors in order of importance. Secondly, an analysis model of STIPAT influencing factors in the BMS is constructed. Multicollinearity is tested, the correlation coefficient is obtained by ridge regression modification. Thirdly, scenario analysis is then employed to set baseline, low-carbon, and high-carbon scenarios to predict and analyze future building CE. Finally, the AdaBoost was applied to test the accuracy of scenario predictions. The results show: (1) The top three influences on CE in BMS are CE per unit of completed area, energy consumption during the BMS, and number of employees in the construction industry. (2) Under the three conditions, CE showed a trend of first increase and then decrease. (3) The error values were all within 2 %, which can be carried to forecast CE in 2045 and after.