Abstract: Strain gauge pressure sensors are widely used in fields such as industrial automation, medical monitoring, and intelligent equipment. However, their output signals are susceptible to ambient temperature drift, which severely limits measurement accuracy in industrial scenarios for precision detection. High-precision demands are not easily met by traditional calibration methods. While accuracy can be improved by common machine learning calibration methods, efficient operation on resource-constrained embedded systems is hindered by their high computational complexity and resource requirements. A novel temperature compensation algorithm based on random forest is proposed in this paper. By introducing interaction terms between temperature and Wheatstone bridge output voltage as feature inputs for the random forest more precise decoupling of temperature effects on pressure measurement results is enabled by the algorithm. Meanwhile, grid search is used to optimize the model parameters and minimize the prediction error. It is demonstrated by experimental results that the maximum relative error over the full scale of the pressure sensor calibrated with the random forest temperature compensation model is 0.13%, indicating higher calibration accuracy than existing temperature compensation models. Furthermore, hardware deployment evaluations confirm that the model satisfies the real-time requirements and resource constraints of embedded systems.