Abstract: Sensitivity analysis of factors affecting soil frost heave rate is one of the most important research contents of artificial frozen soil, which is of great significance to engineering construction. There are many factors influencing the frost heave rate of artificial frozen soil, and those influencing factors are coupled together. The experimental data of sandy loam, clay and loam in Gansu area are selected, and the sparrow search algorithm is used to optimize the back propagation (BP) neural network to establish a predictive model. On this basis, a method for calculating the sensitivity of the frost heave rate affecting factors is proposed. The results show that the prediction accuracy of the optimized prediction model is much higher than that of the traditional BP neural network prediction model. It also reals that the most sensitive factors affecting the frost heave rate in sandy loam, clay, and loam are frost penetration rate, initial dry density in the freezing part and initial water content in the freezing part. This research provides an important reference for preventing engineering frost heave hazards.