In order to explore the influence of relevant parameters on the cooling performance of space-borne pulse tube cryocooler and improve the consistency of cooling performance, a random forest regression model based on machine learning is established to make regression prediction of the cooling performance and various independent variables. The average relative error of cooling performance prediction is 5.62%, and the average certainty coefficient is 0.805. In terms of the influence degree of the variables, the first and second feature are mesh filling rate and magnetic induction intensity, which are consistent with the actual experimental results(the actual input power changes of mesh filling rate and magnetic induction intensity are 6.11 Wac and 3.52 Wac, which are much larger than the other four independent variables). The results show that RFR has the high accuracy and robustness, which provides a new idea for the consistency improvement of the cooling performance of space-borne pulse tube cryocooler.