Abstract: In the construction of deep and thick stratum shafts,the freezing method is often employed,where the unconfined compressive strength is a crucial mechanical parameter in frozen design. Due to the limitations of indoor experiments and the complexity of influencing factors,the applicability of empirical strength formulas is poor. This study utilizes the highly integrated XGBoost algorithm to predict the strength of frozen soil with different particle size distributions. In comparison with other empirical methods,it exhibits higher accuracy. The Pearson correlation coefficient analysis suggested the need for further exploration of the nonlinear correlation between temperature,strain rate,and unconfined compressive strength of frozen soil. The results indicate a strong negative correlation between temperature and unconfined compressive strength;the strength exhibits a rapid increases in the early stage,followed by a moderate increase in the middle stage and a significant increase in the later stage. There is a positive correlation between strain rate and unconfined compressive strength,with varying sensitivities to different magnitudes of the strain rates. With lower strain rates,the strength slightly increases,while with higher strain rates,the strength increase becomes more pronounced. Although different soils exhibit similar trends,variations in particle size distribution lead to differences in final strength. This research provides a scientific basis for predicting soil strength during frozen subway connecting passage construction. 

Key words: frozen method construction,unconfined compressive strength,machine learning,temperature,strain rate