Endometrial cancer (EC) incidence and the associated tumour burden have increased globally. To build a molecular expression prognostic model based on the tumour microenvironment to guide personalised treatment using a machine learning approach. Two datasets were reviewed, including a training cohort (n = 698) and a testing cohort (n = 151). All patients underwent hysterectomy ± adnexectomy ± lymph nodes dissection between December 2014 and June 2020 at the PLA General Hospital First Medical Center and received necessary and regular follow-up. We developed novel models using R software to predict factors that affect survival, such as progression-free survival and overall survival. Then, the model was optimised by evaluating the prediction efficiency in multiple dimensions. Eight hundred and forty-nine patients with EC were included in the study. Survival-related influences on EC patients were identified by univariate analysis and cox regression equations. In addition, a nomogram was visualised in conjunction with demographic characteristics and the above meaningful clinicopathological variables. Ultimately, through a comprehensive assessment, a random forest model (RF16) was developed for complementing the findings of the molecular classification of EC. The RF16 not only specifically characterises tumour molecules, but also enhances the generalizability of the model by replacing gene sequencing with immunohistochemistry. This study showed that the machine learning model (RF16) is low-cost, efficient, and clinically valuable in guiding treatment for EC patients.