Taking the influencing factors of carbon emissions in the construction industry as the clustering index, the provinces with similar characteristics were divided into three clusters. The random forest algorithm was used to identify the key factors affecting the carbon emissions of each type of construction industry. The STIRPAT model was constructed to predict carbon emissions under three different scenarios(baseline, extensive and low-carbon) by adjusting the relevant influencing factors. The results show that there are large differences in carbon emissions among different types of provinces. Category 1 and Category 2 have great potential for carbon emission reduction. It is necessary not only to accelerate the upgrading of industrial structure and energy structure transformation, but also to pay attention to the impact of economy and technology on carbon emissions in the construction industry. The peak time varies. The provinces in Category 1 will reach peaks around 2032, 2032 and 2030 under the baseline, extensive and low-carbon scenarios respectively. Provinces in Category 2 can only reach the peak in 2032 under the low-carbon scenario. Provinces in Category 3 can peak their carbon emissions around 2030 under any scenario.